The bridge is four transistors, and the driver is the board that tells them when to turn on and off. It does that several hundred thousand times a second, in the right order and with the right gaps between.
What it is and why
The driver has four jobs. The first is finding the frequency. You cannot set a Tesla coil's resonant frequency once, because the arc changes it while the coil runs. So the driver measures the current in the tank through a ring called a current transformer and switches in step with it. Current in, gates out, current in again: that loop keeps the bridge on the coil's frequency instead of on one you guessed.
The second is firing early. By the time the driver has decided and the signal has reached the gates, the current has moved on. So the driver switches before the moment it wants, by a lead it is given. Get this wrong in the late direction and transistors die.
The third is deciding when to stop, and there are three separate reasons to. The interrupter, or on a QCW the modulator, says when a bang begins and ends. The current going past a threshold ends it early, because nothing else in a doubly resonant coil limits that current. And the board watches its own supply rail. If the rail sags far enough the driver shuts itself down mid-bang, and from the outside that looks nothing like a supply problem.
The fourth is delivering the charge. A gate is a capacitor, and each transition moves a lump of charge into it and back out again, at the coil's frequency, for as long as the bang lasts. Doing that fast enough is a detail on a short bang. On a machine whose bangs are milliseconds rather than microseconds it stops being one.
The output does not go straight to the gates. It goes through a small transformer, because the top two transistors have their emitters on the switching node, and that node swings by hundreds of volts. You cannot wire logic to a point that moves like that, so the signal crosses magnetically. The first three jobs are done on the board. The fourth is done past its edge, and only the part that holds a gate down when the drive is absent belongs to the driver itself.
What one actually looks like
A small board, a hand's width or so, well away from the power section and connected to it by a few wires. The signal runs across it in one direction:
feedback ring -> burden -> comparator -> logic -> output stage -> GDT -> gates
^
protection ring -> burden -> OCD interrupter -> enable
Dawid Krajcarz drew a board laid out that way on purpose. They publish the driver from their own coil as drobotk/drsstc-driver, KiCad sources under an MIT licence. The sheet is split into labelled boxes that closely match the jobs above: Feedback, UVLO, Main logic, Interrupter, Output, Power. It is not a UD or a clone of one. The credits name Ward, Finn Hammer, Slawinski, Anderson, Goodchild and Gao, and say the design it took most from is a project by Mike on the forum.
Three things on it are good to copy, and one of them disagrees with us.
- The parts doing the deciding are comparators, which is the one decision covered below. A TLV3501 on the feedback, an MCP6567 on the overcurrent and another section of it on the undervoltage. The 74AHC14s are on the fibre receiver and in the logic, where a wandering threshold costs nothing.
- The lead is a knob rather than a part or a number.
R120R in series withRV1, a 100R potentiometer, and a fixed 4.7 µHL1, described in that board's own notes as "Phase lead adjustable with a potentiometer, no need for a tunable inductor". The UD line fixes the lead as an inductor you fit, and the UD3 sets it as a time in nanoseconds. A pot is a third way, and the cheapest of the three to change. The same list gives a reason for needing less lead, "74AHC high speed logic, less phase lead needed". Faster logic means less propagation delay in the chain, so less of the lead budget goes on it. - It takes one ring, not two. The first line of the feature list is "Single CT input for both clock and overcurrent detection", and the sheet shows it:
J1markedFB CT, one 5R1 burden, and that same node running to both comparators. We recommend two rings, below. The drawing shows that a single input is a choice a designer can make on purpose in 2026, knowing the tradeoff, and the UD3 makes the same choice. It is not just a corner cut by someone who had not thought about it.
We would feed it from two rings. The first says where the current is, for the frequency and the lead. The second says how big it is, for the protection. Taking both off the same ring is tempting and it is the wrong economy, for a reason worked through on the ring's own page.
The interrupter should come in on fibre. A copper input is a wire from your low voltage logic into a machine that is radiating hard. The fibre is there for isolation, not only for noise immunity. What it guarantees, and the rule that a broken link has to mean off rather than on, are in the fibre and the fail-safe rule.
The frequency the driver chases is not a single frequency. Coupled to a secondary, the pair has two resonances rather than one, and which one the driver lands on is somebody's decision, not an accident. That is covered in there is no secondary frequency. On a ramped machine the frequency also moves while the bang runs: 390 kHz down to 330 on one published coil and 413 to 377 on another (the frequency slides down the ramp). With tens of kilohertz of movement inside a single bang, the first job cannot be done with a number you set in advance.
Steve Ward's open UD line is the reference every other driver gets compared against, whether a builder runs one, a clone, or something of their own described by how it differs. He published the design on his own site in 2008. The revision with a schematic, board files and a bill of materials published against it is UD2.7, on loneoceans. Daniel Marks rebuilt that revision through-hole as UD2.9, in the DRSSTC PCB Pack and under CC-BY-SA. Waska Labs carries the line on as UD29X, Phillip Slawinski's UD+ replaces the discrete logic with a CPLD, and Netzpfuscher's UD3 replaces it with a processor. The UD3 is the board we keep using as the example of what a shipped default can cost you.
The revision you are holding changes the part numbers but not the logic. That matters the first time you follow somebody else's modification (the UD boards). It matters again when you take a board's shipped settings as a starting point. Those were typed for whatever coil the author had, and a programmable board will report back the number you set whether or not the hardware behind it took it.
The one decision this page is enough to make
Whether the parts doing the deciding are comparators.
A Schmitt trigger's thresholds wander with temperature and supply voltage. The driver needs a predictable threshold in two places, deciding when the current crossed zero and deciding when it is too high, so both want proper comparators. A 74HC14 there can work, but you cannot count on it.
Everything else (the lead network, the feedback strength, the protection behaviour) comes with a number attached, and those are covered inside the driver.
What will get you
How to probe one of these boards anyway, and when you need a differential probe, is covered with the scope.
What goes wrong
- The coil will not start, or runs at a frequency nobody asked for. Feedback. The current transformer's winding direction is the ten-second check.
- Transistors die at one tuning and survive at another. The lead, in the late direction.
- The protection never fires. It is a threshold, not a knob position, and it is not a setting.
- A board that worked stops the moment a probe touches it. See above.
- The bang starts correctly and dies partway in, at the same point every time. This is usually the driver's own supply rail sagging into its undervoltage lockout while the bang runs, not the protection or the feedback. The gate waveform captured at the start of a bang looks perfect, so it is easy to miss. Put the scope on the rail rather than on the gate, for the whole length of a bang.
- It runs, but at a frequency that is not the one you tuned for. There are two to land on and the driver picked the other, which is not a fault but a choice somebody did not make.
- The gates look nothing like squares. That is the loop past the board's edge, not the board. See reading a gate waveform for what the shape is telling you, and the gate loop is an RLC for the resistor that sets it.
- The first cycles of every bang are wrong and the rest are right. That is expected. The loop has nothing to follow until there is current in the tank, so every bang starts open loop.
- The driver dies when an arc hits something earthed. This is a different failure from all of the above and it does not start at the driver. See what a ground strike does to a DRSSTC.
Where next
- Inside the driver, the same board with the numbers.
- What a phase lead actually buys, the lead the second job above is given.
- Switching at zero is the failure case, because the obvious target for that lead is the wrong one.
- Scoping a coil without killing it, for the warning above and the rest of what a ground clip does.
- Converting a DRSSTC driver, what has to change on one of these boards to run a ramp.
The driver is the box at the bottom of every department diagram, with the gate transformer above it and the current transformers feeding it from the right.