Trying to thin out the collection before I move out of state!  Buyer pays shipping, UPS, USPS, FEDEX.  I accept paypal, PM me if interested! 

I'm currently "reforming" all units listed below, if any show high leakage i'll remove them from the listing.  I suspect, however, that they still have good life left in them for experimental use.  I set the price, hopefully, low enough to entice buyers as I'm moving out of state next month.  I'd entertain offers if someone wants them all for a deal or is broke but really wants caps for their coil.  I have too many caps and these need to go!

43 types available (from left to right in picture, dimensions do not include terminal post):

Nippon Chemi-con 10,000uF 400VDC, 1620grams, 3.5" x 6.75", QTY 4 available at $20 each (Ive used these within the last few years, light duty filter applications)

AEG 12,000uF 350VDC, 1480grams, 3" x 8.7", QTY 2 available at $10 each (these might be from my DRSSTC2, or they were spares, i cant recall, but they're old-ish).

ADz (?) 22,000uF 200VDC, 985grams, 3" x 5.6", QTY 2 available at $5 each (these might be from my DRSSTC1, could be good for 120VAC doubler for powerful coil).

Nippon ChemiCon 15,000uF 350VDC, 1620grams, 3.375" x 7.4", QTY 2 available at $10 each, one has a dent but still seems to hold a charge fine. (oldish, backups for my big QCW FAT coil). 

I've been on somewhat of a 3-phase kick in my career, focusing primarily on multi-phase permanent magnet motors and inverters. I had, long ago, seen a picture of a set of 3 SSTCs, by experimenter Duane Bylund:

http://www.duanesradios.info/html/tesla_coil_photos.html

It was 3 of these units, with their bridge switching synchronized and phase shifted, which is essentially the approach with my 3 phase machine. 

Cutting to the chase, you can watch a few videos of this little monster:


The power stage is 3 half-bridges.  With only a half-bridge feeding each Tesla coil through a "larger than resonant" (LTR) primary capacitor, power is controlled/modulated by  excitation frequency, particularly above the secondary resonance (upper pole freq).  By switching further above resonance, power is reduced, and switches of the mosfet type, operate essentially in zero-volt turn-off (ZVS) mode with low switching losses. This is because the mosfet can switch off quickly - limited mainly by gate driver - but the voltage rises as a function of the total output charge of the half-bridge and the current at the output during the transition.  So the voltage across the mosfet junction while it transitions from on to off is less than the total supply voltage, and switching losses are reduced.  The truth is, the switch off loss is still significant at the operating frequency range of 540khz to 375khz.

With the primary capacitor set "very much larger than resonant", each of the 3 coils looks electrically like the classic SSTC with a "dc blocking" capacitor on the primary side.  A SSTC self-oscillates at a frequency above the secondary resonance, just like a QCW-DRSSTC, at the "upper pole" frequency.  Because the "dc blocking" cap sets the primary side resonance to a frequency far below the operating frequency, its typically ignored, however the interaction of the mutual inductance and the secondary self inductance fields are still "fighting" each other so that the effective inductance is less than the secondary on its own and so the system oscillates at f>Fsec.   In my system the primary side cap is only marginally LTR so Fpri < Fsec, initially, but ideally Fpri = Fsec_loaded.  This condition, i think, minimizes stored energy (lowers Q) which is considered extra "cost" as it doesn't directly contribute to the energy in the spark.   In practice, Fpri < Fsec_loaded in my setup because I still wanted to limit the power as the inverter comes to switch near ZCS.  I find this is a practical compromise when its not easy to re-wind the primary with more turns to raise the impedance.  Also, keeping the primary a little detuned helps linearize the power vs frequency transfer function.  You can think of it as just partial "power factor correction" by nearly cancelling the primary leakage inductance with the series cap.

The first experiments utilized a semi-retired prototype from work which featured GanSystems 650V mosfets, 12 of their largest 25mOhm fets ganged up in parallel per switch location for ultra low Rds and reasonable charge losses for motor inverters (and softer switching for ZCS/ZVS tesla coil driving as the junction charge is relatively benign if not slightly helpful).  This early stage experimentation was just a hacked version of a motor controller I'm developing.  It was also lack luster performance because I was too shy to stress the inverter as it was particularly expensive in terms of part cost.   

After the GaN switches, I tried some of the new 4-lead TO-247 style Si MOSFETs from fairchild.  This showed a lot of promise up until its ultimate demise, of which im not sure what the cause actually was.  I suspect a bug in my controller allowed switching below the resonant (zero phase) frequency of the system, giving hard body diode recovery at voltage slew rates outside of the SOA triggering their prompt failure.  Disappointingly, all 3 half-bridges failed so energetically that they were a complete loss, recovering only the heatsinks for future use. 

I decided to move on with SiC devices instead, figuring they should be perfectly fine with body diode recovery if it happens, but also their lower gate charge makes switching them faster, more realistic.  Also, they are widely available in the 4-lead TO-247 design that i had my boards laid out for.  I chose the C3M0021120K devices, but if i were to re-do it, i would consider the of United SiC cascode mosfet, especially their new 650V parts in a TO-247-4L.  The switching losses seem to go up disproportionately for the 1200V devices, but its hard to resist the extra power density if you can still dissipate the heat. 

To help with heat removal, I've chosen a "live heatsink" design so that each device has 1/2 of a small CPU cooler directly mounted to it with a small amount of thermal grease, but no isolation pad which usually has 3-4X the thermal resistance as the internal resistance of a high-end 247 device.  Direct heatsinking alone should result in significantly lower junction-sink temperature difference.

Gate driving is handled by a simple boot-strap type driver.  Gate drive transformers simply have too much leakage inductance and also supply inappropriate negative drive voltage for Wide Band Gap (WBG) semiconductors like SiC or GaN (with some exceptions like cascode SiC which actually has a Si mosfet control input).  My design uses a 10A gate driver IC (NCP81074B) located immediately next to the G-S pins on the TO-247-4L fets.  The 10A driver ICs are fed isolated switching signals from yet another gate driver IC which has built in dead time generation (Si8231).  There exist better options for driver ICs and isolation if taking a fresh start, but the architecture would look mostly the same, locating the driver ICs as close as possible to the mosfets, which in my case required individual drivers for high and low side fets.

The switch timing is generated by a STM32F405 microcontroller, using its timer/counter peripheral which has 3 phase output capability.  The 3 square wave drive signals for the 3 half-bridges are phase shifted by 120 degrees (or close to it, depending on quantization limits of the timer/counters).  The primary current of each coil is monitored by a CT, checking the current amplitude with a ADC, and also checking the zero cross timing with a high speed comparator feeding into a timer-capture peripheral that is synchronized with the gate switching signal.  The MCU is constantly watching how much "phase margin" there is between mosfet switching and primary current zero crossing.  The goal is to never switch later than the current zero crossing, and ideally keep switchings slightly ahead of it so that switching losses are reduced by ZVS conditions.  While the MCU approach cannot keep up with every zero crossing at 500khz, it only has to "spot check" the phase margin every few cycles (at 20uS update period) because of the quasi-steady-state nature of the system.  That is to say, there should be no fast changes to track, and we can generally assume its "safe" to raise the driving frequency if necessary as it will reduce the power throughput and also give safe switching conditions (no body diode forced recovery). 

The remote control "ramp generator" is essentially commanding the operating frequency of the inverter.  Lowest power sets the frequency to 475khz, while the highest power setting is 375khz, provided the zero cross detector sees enough phase margin for near ZCS, and also that the ADC reports primary current within limits.  If the commanded frequency is "too low", the controller will just track the coil's natural oscillation frequency via zero cross measurements and slow adjustments of the timer/counter period.  As the loaded coil frequency drops, the primary tuning boosts the power, helping feed the sparks.  The ramp generator i use features a 5-point piece-wise linear waveform generator so that i can experiment with shaping the ramp function to optimize spark growth.  While this functionality proved useful for a single coil making long sword sparks, it doesn't really matter much for the chaotic, short sparks created by the 3-phase setup.

The start-up of the system works at a much higher frequency still, 540khz.  This is to gracefully bring the system into stable oscillations without seeing the "transient" response that would otherwise show up when driving closer to resonance, which generates too much voltage for the coils to stand.  For the first 400uS or so, the drive is sweeping from 540khz down to 475khz.  At 475khz plasma is just starting to form, which quickly brings down Qsec and brings stability to the system.  From here the ramp generator is setting the frequency command, and the zero cross detector is checking if theres enough phase margin to allow it (generally, if there is not, it means the spark load has not grown enough, but given time it will usually get there).

The machine has a built-in active rectifier (PFC).  The PFC includes a number of nice features.  Inrush/pre-charging is handled by the full-wave bridge rectifier common for a single-phase PFC.  The bridge rectifier uses SCRs for the two "upper" diodes feeding the positive voltage input to the boost circuit.  The SCRs are phase-angle controlled by the microcontroller (which is also watching AC line zero crossings).  Therefore, no external resistive/capacitive charging element is needed, and no extra relays/contactors are needed.  The pre-charging profile could be fine tuned to maximize charging current, but i just use a linear adjustment of phase firing angle over time until reaching full 180* conduction, at which point the PFC boost action can begin.  The precharge current pulses are on the order of 15A peak.

Since the PFC is controlled via MCU, it is easy to build in voltage setpoint (380V to 800VDC), which is a slider on my remote control box.  I decided not to bother with variable input current setting, however this would be easy to add.  Instead, if the MCU detects 120V line level, it sets the PFC max current to 15A RMS, but if 208/240V line is detected it allows up to about 57A RMS, or something over 13kW output from a very compact design weighing maybe 5lbs excluding the large output capacitors.  Of course, i can't claim it could reliably operate at this power level continuously, it probably would not be very reliable, but it seems to work OK for short term duty cycles.

PFC current regulation is also handled inside the MCU.  Once per PWM cycle (at 50khz) the line voltage is measured, the PFC boost current is measured, and a new duty cycle is calculated to maintain sinusoidal input current.  The voltage control loop, which drives the current setpoint, updates at every AC line zero crossing to avoid distorting the AC line current.  This is similar to what an analog PFC chip would do by low pass filtering the voltage loop to avoid any response at line frequency. 

Speaking of power... one significant issue was over-heating of the secondary coils.  Eventually i added an acrylic "chimney" around the coils and used 150mm fans to blow air up past the primary MMC, through the primary winding, along the secondary coil, and venting out of the toroid (which can also get warm!).  I have a few destroyed coils as the PVC form expands from the heat, stretches the copper wire and then retracts, leaving the wire loosely hanging on the form.  The fan cooling seems to handle this problem, just fine.

I plan on taking a deeper dive into more details of the design and operation of this machine, this was already getting to be a long introductory post.  Please note that some of the pictures are a little outdated with respect to current hardware (like, the half-bridges shown are my Si versions but the SiC version looks almost the same).

I also have plans to power this setup with a 444V lithium ion battery, capable of supplying enough current to run all 3 phases CW.  More to come...