Powered by:  Fairchild International Rectifier On-Semiconductors Maxim-Dallas Texas Instruments Microchip


BJT flyback, MOSFET flyback, MOSFET half-bridge I, II, audiomodulation, IGBT

      Another way how to drive TC is using semiconductor parts. Usually the bridge circuit with power MOSFETs is used. But semiconductors are so fragile. They dislike (short) high-voltage/current peaks. And the capacitive current to gate is quite big at frequencies about 1 MHz. Gate capacity of usual power MOSFETs is about 2 - 10 nF. This means that we need power gate driving circuit which is able to source/sink several amps to/from gate electrode during peaks. They are manufactured as integrated circuits but it's hard to buy them in Czech Republic. Sometimes when I'm buying electronic parts it seems that I'm living in Africa ;-). Our professors on our eltech. faculty (of Czech Technical University) show us (in subject Theory of electromagnetic field 1) a small TC powered by single transistor self-oscillating driver. Thanks to prof. Pankrác who allowed me to explore the SSTC driver PCB and draw schematic - here is it:

ČVUT singleT driver

SU169 is high-voltage switching transistor from deflection circuit used in old B/W TV. Its max. parameters are: UCB0 = 1000 V, UCE0 = 400V, IC = 10 A, UEB0 = 8V, Ptot = 100 W, h21E >=15. Diodes may be some generic rectifier 1000V / 1A, faster are better. I built it at home on a small PCB and it worked on first light. It draws about 1,5 A at 35 VDC and do about 2 cm sparks with quiet hissing noise like HV from TV. When I applied higher voltage to driver the sparks become longer but only for a while. After few second the transistor "pops" breakdown and game over. I tried to use other transistors such as SU161, BU208... but they died too. Usually the junction B-E was shorted and junction B-C blown up. I have only old oscilloscope Křižík (assembly date about 1950 :-) which can't displays short voltage peaks or other evils damaging my transistors. Here is some photos of my SSTC driver on PCB.

SSTC driver front SSTC driver front-left SSTC driver back

And here how it works:

SSTC sparking SSTC small glowing tube SSTC variator SSTC compact glowing tube SSTC glowing tube

MOSFET flyback

      26.9.2002 I was thinking about usage of power MOSFETs to driving tesla coil for a long time. Bridge circuits seems too complicated to me. This is because driving a high-capacity FET's gates is not easy at high frequencies about 1 MHz. I have an idea to replace vacuum tube by FET in my VTTC flyback driver. I found an IRF640 (200 V; 18 A; 125 W; 0,18 Ω) FET in my stock. Voltage from feedback coil is clamped with two Zener-diodes KZZ75 to protect gate (I know that it is not optimal solution). I powered the circuit from variac (half-wave rectified and filtered by 100 µF capacitor). To start the circuit oscillating a short connection of gate on positive voltage via hi-ohm resistor is needed. Then I connected resistor to ground to prevent FET damage if oscillations drops out and FET would remains opened. This will need better solution. When I powered it by smoothed 70 VDC I got sparks like with SU169 BJT driver (on my big TC). Then I reconnected it to variac half-wave rectified output. This made sparks like PL504 VTTC driver at about 120 VAC. And there's no serious cooling problem. When I applied higher voltage FET popped. I'm planing parallel connection of about four IRF740 or IRF840 powered directly by rectified mains. It will simplify the power supply for the driver and make it smaller.

FETTC scheme

500VA variac FETTC sparks FETTC sparks

MOSFET half-bridge I

      14.5.2003 After IRF640 has died in my MOSFET flyback I tried to replace it by parallel combination of IRF740 and other MOSFETs but with no result. The circuit won't start to oscillate in any case (nor with IRF640 originally used). I checked and replaced everything that I can but the circuit was never started oscillating like before.
      I successfully obtained two samples of MAX5048 integrated circuit via Internet. It has especially developed for driving power MOSFETs. It's incredible how this IC in tiny SMD package can handle 7,2 A peak current to gate. I tried to drive a common 5 nF capacitor. At 1 MHz the waveform was still looking quite good but circuit and capacitor was heating. With this ICs I finally could start to build my own MOSFET half-bridge. The half-bridge behaves like an electronic alteration switch which alternately connects the load to +Ucc rail and ground rail. The major advantage of half-bridge compared to flyback is that half-bridge has better efficiency and limits Uds across MOSFETs to maximum of +Ucc rail voltage. It has also a disadvantage - danger of simultaneous MOSFETs conducting called cross-conduction or shoot-through condition which cause immediate MOSFETs damage.

half-bridge principle

      The first problem needed to be solved was driving the high-side MOSFET. It's gate floats at high voltage level and cannot be connected to MAX IC directly. So I decided to use insulating transformer called GDT - Gate Driver Transformer. The GDT requires high quality magnetic core (low RF losses by eddy currents). Also it needs to be trifilar wounded to get high coupling coefficient. It has limited bandwidth excluding DC and too low or high frequencies. And output DC bias depends on input duty cycle. But this doesn't matter when we use it for driving a tesla coil which has very narrow bandwidth and needs 50% duty cycle. In other hand there's a lot of advantages:
1) galvanic insulation of GDT windings (also protects the driver when MOSFETs fails)
2) easy phase shift of 180° by simple swap of winding leads
3) GDT may have more windings (one GDT for full-bridge)
4) transformation ratio can be other than 1:1
5) when any driving signal present both MOSFETs have shorted gates and are closed
6) zero DC bias with 50% duty cycle cause symmetrical ±gate voltage (gain noise immunity, reduction of cross-conduction while switching).
More details how to make a good GDT you will find in this design guide.
      Now I try to explain why so big gate currents are needed for power MOSFET's driving. FET means Field Effected Transistor - transistor driven by (electric) field. This evoke impression that there's no currents and no energy is needed for driving but this is a mistake. Any chick paws for free ;-). By principle of MOSFET there's always some capacitance between gate-source and gate-drain. And this capacitance is not insignificant small in power MOSFETs. IRF640 has Cg about 2,5 nF and total gate charge 67 nC. On every turn on/off the Cg has to be charged/discharged from supply of driving signal. This effect is insignificant while switching frequency is low. But I need to operate my half-bridge at 1,2 MHz (0,8 µs period). For reasonable losses MOSFETs should switch 10-times faster. It requires that signal supply, with output impedance R, charges the gate in tau = 0,08 µs up to 10 V. So R value must be about R = tau / Cg = 30 Ω. When gate is totally discharged it behaves like short circuit and through 30Ω resistor will flow 0,5 A peak current from 15 V supply. Of course the RMS current is much smaller. For faster MOSFET switch on is lower R and higher peak current to gate needed. Semiconductor manufacturers offers integrated drivers with peak current up to 12 A and switching times about 40 ns nowadays. The capacity gate-drain is significantly smaller but more problematical. It causes Miller's effect - transmission from output to input. When lower MOSFET turn off quickly voltage on its drain rises quickly to +Ucc rail voltage. This make a pulse which is transfered via mentioned capacitance back to gate. It may cause MOSFET's turn on again or its damage (common power MOSFETs have max. Ugs = ±20V). Therefore we need signal supply with low output impedance which minimize influence of reverse transmission. Also it's good idea to put some limiter between gate and source like two antiserial wired Zener diodes which clamps Ugs to max. ±Zener voltage of this diodes and protects MOSFETs.

disabling MOSFET body diode Another problem is how to eliminate MOSFET body diode - it is a diode "connected" between drain and source (in N-channel anode to source, in P-channel cathode to source; usually not drawn in schematic sign). This diode, which we got for a free from manufacturer, is given by manufacturing technology and cannot be removed. It's useful as free wheeling diode for low frequency application but causes problems at higher frequencies. It's reverse recovery time is much times longer than switching times of MOSFET itself. For example IRF640 body diode has trr = 170 ns and MOSFET's turn off time (tfall + toffdly) is 30 ns. We can eliminate the body diode when we don't allow to open it. (When inductive load is turned off there always flow some current back to supply; we must make some path for it - free wheeling diodes instead it damages our MOSFETs). For this purpose we connect fast Schottky diode between drain and source. Schottky diode has lower threshold voltage and opens sooner than body-diode could. But Schottky diodes usually have low reverse voltage allowed. So we connect some common diode in series with MOSFET which increases total forward voltage droop and then we connect a fast diode parallel to this combination as is shown on left picture. If we didn't eliminate the body diode it causes that MOSFET stays opened after turn off due to long body-diode trr. When opposite MOSFET turns on it make a shorted circuit. Big (cross)current will flow from +Ucc rail to ground via both MOSFETs and it may shoot them out of your board. The SSTC theory is very detailed described on Richie Burnett's website. I recommend to read it.

half-bridge scheme with MAX5048 driver

      Here I used two MAX5048 drivers which works in full-bridge. This gives me ±12V voltage amplitude on secondary side of GDT with 1:1 transformation ratio. I could use only one driver and GDT with 1:2 transformation ratio but this will 4-times increases the output impedance (impedances are transformed via square of transformation ratio). GDT primary is DC separated by capacitor. It's not needed for 50% duty cycle when DC bias is zero. But it's very useful for drivers protection when clock signal is lost. If it would happen one driver's output stays at High and other at Low and flowing DC current will damage them. I recommend to fuse the drivers voltage supply to protect them. 75451 circuit works like TTL->CMOS converter and "deadtime controller". It's purpose is prevent to overlay both driving signal using very shot delay while both signals are Low and both MOSFETs are in off state. Using capacitor-trimmer you can vary the delay in some tens of nanoseconds. In practical view the deadtime doesn't approved after connecting MOSFETs. Gate voltage curve has been smoothed. So I don't know may be using of inverting and non-inverting MAX's input to generate both signals would be good enough.
      For synchronization with TC I used PLL circuit 74HCT4046. It is set (via R1, R2, C1) to generate frequency around 1,2 MHz without any input signal. After TC is partially excited at it's natural resonating frequency it induces some feedback signal to antenna of PLL circuit. PLL will lock at this frequency. When secondary frequency changes PLL will automatically retune to serve right frequency. Input of PLL is protected with two clamping diodes against too high input voltage. As an antenna I use few centimeters of wire or connection at one lead of feedback coil via resistor.

PLL scheme

      Here are some oscillograms showing waveforms in interesting nodes. I used for this job my new two channel oscilloscope Grundig MO 52 (it really couldn't be measured on my old Křižík scope from 50' :-). Time resolution is 0,2 µs/cm in all graphs.

Ug1, Ug2 Uout, Icross Uout, Uin Uout, Ipri
Ug1, Ug2 Uout, Icross Uout, Uin Uout, Ipri

On 1st oscillogram you can see driving voltage at gates of both MOSFETs. This is nothing beautiful but designing a driver at 1,2 MHz is a real nightmare. I tested it with simple generator which cannot give 50% duty cycle. On 2nd one is displayed output voltage (which looks quite well) from half-bridge and supply current (with no load). So some cross-current is flowing. On 3rd one is output voltage from half-bridge and input voltage captured by antenna at PLL input. And on 4th one is current flowing to primary coil and half-bridge output voltage. There's some ringing due to long wires which I used for testing.
      I tested with 70 VDC power supply and I got 3 - 4 cm spark length. Sparks was very hot and melted my needle discharge electrode.

winding trifilar GDT FET half-bridge workplace FET half-bridge spark

MOSFET half-bridge II

      23.5.2003 So I burned whole driver board with my "skilful" :-(. By a mistake I swapped 12 V with 35 V supplying in a chaos of wires from my power supply. Tantalum SMD blocking capacitors was sticked into my table and MAX ICs was well damaged too.
      When I explored Internet I found some better MOSFET driver circuit - Texas Instrument UCC27322, which gives 9A peak gate current. BTW its due to unique technology of combination bipolar and MOS transistors in output stage. MOS is faster and bipolar have lower Uce drop in turn-on state and this lowers power loss. I decided to design my circuit and PCB from zero. 1st I made small comparison of last MAX5048 with new UCC27322. Both circuits had to drive 10 nF capacitor at 1 MHz. Oscillograms shows that UCC27322 produce faster rise time (time resolution is 0,2 µs / cm). The circuit was made simply on test-board so longer connections caused some ringing. I expect that on PCB it will be better.

UCC27322 SMD package on my finger UCC27322 SMD package bottom MAX5048 loading 10nF at 1MHz UCC27322 loading 10nF at 1MHz
SMD package heatspreader MAX5048-10nF UCC27332-10nF

UCC27322 IC have interesting metal facet at bottom of its package. Silicon chip is directly seated on it. It helps with heat sinking out from a chip. When proper cooling plate is used UCC27322 can withstand 1,3 W power dissipation (compared to 0,5 W of DIP-8 package)! So I tried to make large cooling copper plates and short power paths on new PCB layout. I placed SMD tantalum and ceramic blocking capacitors everywhere. Because driver IC heats too much at 1,2 MHz I reinforced cooling with copper plate pieces soldered on PCB. Also I use only 12 V instead allowed 15 V.
      Then I rewound GDT trifilar 3x16 turns on new high quality ferrit core Amidon FT 82-43 (d1 = 21 mm, d2 = 13 mm, h = 6,4 mm, Bmax = 0,275 T, µr = 850 - 3000 do 30 MHz) which I bought in GES Electonic for 90Kč (about 2,6 euro). (There's still old GDT on PCB photo).

PCB bottom PCB top winding new GDT

      Here is the schematic:
half-bridge scheme with UCC27322 driver

      Function of this circuit is very similar to old circuit. I improved deadtime control. Now four Schmitt inverters are used which has small own delay serving minimal deadtime. And they are shaping integrated edge of square wave after RC circuit at NOR gate input (which is not Shmitt-like). Coming CLK signal with High level is divided. One component go directly to input of NOR gate. Second component is delayed first and then go to other NOR gate input. NOR gate output goes to Low immediately because only one active input is enough for it - see the truth table below. After roll over CLK signal to Low there's still second NOR gate input held on High due to RC delay circuit for some time and NOR gate output stay at Low. After RC delay elapsed (30 - 80 ns) NOR gate output turns High. This cause that Low period is little bit longer than High period. The same situation is in other driver working with inverted CLK signal. This mean that there cannot be overlay between roll over of both signals.

 A   B   non A   OR   NOR   AND   NAND 
0 0 1 0 1 0 1
0 1 1 1 0 0 1
1 0 0 1 0 0 1
1 1 0 1 0 1 0

Uout UCC27322 Uout GDT Uout GDT-FETs Uout, Iout
Uout UCC27322 Uout GDT Uout GDT-FETs Uout, Iout

      On 1st oscillogram can be seen the output voltage of both UCC27322 with no GDT connected. (0,5 µs / cm). On 2nd one is GDT secondary voltage with no MOSFETs connected (0,2 µs / cm), on 3rd one with connected MOSFETs. On 4th one is half-bridge output voltage and TC primary current. They should be in phase with proper tune. When MOSFETs will switching in time while current passes zero (soft-switching) it would lower switching losses significantly. I didn't tune it yet. Then I found that connecting TC primary via capacitor which bring it to resonance is better than connecting it to capacitive divider with relative high capacitances.
      1st I tested it with 35 VDC and 70 VDC power supply and then with variac (half-wave rectified, 100µF filtered). I got about 8 - 9 cm spark length with 150 V input voltage and 2,5 A current draw. Sparks was very hot. When I tried to pull arc with screwdriver from secondary MOSFETs had blown. I damage them a lot in further experiments. It is not cheap entertainment. It usually has shorted all three electrodes and driver circuit drives this short. So I insert 400mA fast fuse into +12 V line to protect them. Problems also occurred when I supply the driver from unfiltered power supply. There's some problems with PLL synchronization. Maybe its filter or what? When circuit is out of tune the MOSFETs are stressed much more. Or simply 1,2 MHz is too much for this electronic parts (hoping better future). I don't know anybody who success at this frequencies. I saw this working at max. 400 kHz. But this wouldn't discourage me. Maybe I will wound some bigger TC with lower fres...

hot sparks hot sparks hot sparks

      29.8.2003 I hadn't described details of connecting TC to half-bridge above. So I would set it right now. There are three choices:

connecting load to half-bridge

The first is interesting in that it doesn't use any primary coil. The output voltage from half-bridge is directly feed to secondary base (base feed method). This have some advantages that we don't care about number of primary turns and coupling. Also there is reduced flashover problem. And circuit is easier to analysis, see the theory. Major disadvantage is relatively high impedance at resonance. It needs several kilovolts to be employed for driving and this isn't easy to be done with nowadays MOSFETs. And the coil is galvanic connected to half-bridge power supply, usually mains, that is dangerous.
      In the middle is classic circuit (I have never seen anything else on foreign websites) which using non-resonating primary coil. Second winding terminal is connected to the node of capacitive divider which removes DC bias and makes RF filtering on power supply rails. Due to transformation effect the impedance in resonance is much lower than in first case. Secondary winding is naturally insulated from half-bridge power supply.
      The third circuit using resonating primary coil. There's a Cres capacitor in series with primary winding. Its value is calculated to bring primary in resonance at same frequency as secondary fres. Then the impedance is even lower than in second case. It allows to use lower voltage power supply or increase sparking with current power supply (with enough powerful MOSFETs). It may interests foreigners from countries with 120V mains.

      16.9.2003 Finally I decided to wind a new bigger TC with lower self-resonating frequency. I hope to have less problems with semiconductors switching at lower freq. More about BigTC here. Now I have TC with fres about 330 kHz. So I can buy more powerful MOSFETs now: IRFP460 (500 V, 20 A, 250 W, 0,22 R). It had bigger gate capacitance of course. I think that my UCC27322 driving ICs would burned at 1,2 MHz with this MOSFETs. But at 300 kHz are acceptable warm. New MOSFETs seems to be very robust because I didn't burned them yet (knocking on the desk ;-). I also replaced serial diodes in MOSFETs sources with more powerful Schottky diodes SR3060 (2 x 30 A).
      Unfortunately there arose one big problem with PLL circuit. As I was warned at HV-forum the xx4046 PLL's 1st comparator makes 90° phase shift when locked properly! It explains mentioned phase shift on the previous oscillograms. So I tried to use both remaining comparators in PLL IC but with no success. This comparators sets VCO to minimum frequency when input signal is not present. I set 260 kHz. I also limited maximum frequency to 450 kHz by R-C network. The problem is that PLL is refusing to lock while I'm doing with it everything I can. It produces fmin or fmax or locks at some wrong frequency. If it occasionally locks (usually at high power supply voltage) at right frequency the lock is very unstable and several frequency jitter can be seen. Also the phase is not perfect. Sound of sparks is strange squeaking and random changing. When power supply voltage is significantly risen or lowered PLL unlocks. I tried to insert an inverter between PLL and MOSFET driver board. I also tried to pickup feedback signal from current transformer at secondary base instead from antenna but with no result.
      I don't know why the first PLL's comparator lock so easily as well as with low power supply voltage and other two not. I tried to make a bizarre circuit with two PLLs in cascade. I hope that both 90° phase shifts would add to 180° and then I make a 360° shift by inverter. But it works only for very low power supply voltage. With voltage rising the phase shift fast moved to about 90°.
      In this moment I have no idea how to solve this. I'm asking you for help if you have some experiences with xx4046 ICs or other PLLs.

2 PLLs in cascade Uprimary vs Usecondary Uprimary vs Usecondary
2 PLLs in cascade Upri vs Usec Upri vs Usec

      13.8.2004 I finally have a time to update my TC website today. I will describe my advances with SSTC mostly from this year's winter time. Betweentimes my SSTC works quite well for a few months. It reached streamers up to 38 cm long (running from 230 VAC mains). But read in sequence...
      After many experiments with my current half-bridge PCB it seems to me proof enough so I decided to make some nice housing for it. I don't like many things lying around because it increases probability that something would accidentally touch together and blow up along time (experience by my own :-). I also put an emphasis on good cooling and RF shielding. The base was made of 3 mm thick "L" shaped aluminium plate (bottom and one side) which holds black profile of heatsink 12 x 10,5 x 3 cm making another side. Opposite side holding control circuit is made of the same plate. The remaining side is made of alufoil-coated (the inner side) wooden board. The top cover is not done yet because I often need to have access inside. The PCB is mounted on spacers at bottom side of the box. Power transistors are oriented to face the heatsink side. To minimize the thermal resistance between transistor's case and heatsink I used a kapton insulation pad that I bought in GES. The pad is only 0,05 mm thick but it have dielectric strength 3,9 kV and thermal resistance cca 0,53 K/W (for a TO-247 case). It's better than common silicone pad which has Rth cca 0,7 - 1 K/W. Even better are insulation pads made of Al2O3 ceramic that can reach Rth down to 0,19 K/W at 1 mm thickness. To eliminate micro air gaps between transistor case and heatsink I applied white silicone thermal-conductive pasta. On opposite side of transistor's case I mounted a small aluminium "U" profile to improve cooling (like a sandwich) but most of heat is transferred through metal back side. Anyway it may be a good protection against the transistor's case splinters during its explosion :)
      Because of increasing the driver power I also replaced diode network around MOSFETs using FYAF3004DN in series (fast Schottky diode 2 x 30A / 40 V in insulated TO-3PF case) and RHRP1560 parallel to MOSFET (hyperfast diode 15 A / 600 V, trr = 35 ns in TO-220 case). I mounted the diodes on small heatsink mounted on bottom side of the box. Then I built in a linear stabilized power supply of +5 V and +12 V for gate drivers and support circuits. To avoid a need of using variac I made a thyristor pre-regulator modulating rectified mains voltage. It has also switched banks of filtering capacitors 220 µF and 390 µF that may be used in any (parallel) combination or disabled. Mains voltage is filtered through a line filter made of toroid ferrite choke and capacitor. Here's the result of my effort:

chassis chassis chassis chassis

      The thyristor pre-regulator uses 2N6509 (BT151 800R before) in classic phase controlled rectifier circuit. There is also a single forward biased diode in series with thyristor that should stop a negative voltage if thyristor would fail. The thyristor can be bypassed by a switch so then it will work as a simple half-wave rectifier. Heating of thyristor during operation is minimal. This thyristor circuit also brings one big problem which is implied from the principle of operation: the thyristor is turned on with 0 - 180° (in my case 30 - 170°) phase delay in very short time. When the phase delay is set to 90° there is extreme dV/dt - voltage rise from 0 to +320 V within a few microseconds! But this is not acceptable for electrolytic filtering capacitors and even the PLL circuit is unable to lock so fast. When I tested it first time I had attached a 10&micr;F elyt capacitor and it became hot during very short time of operation. I catch up the power button in right time before it could explode (I know that exploding capacitors are funny but it smells like a piss and made paper rubbish around so I prefer to do it at school instead at home ;-). When a bigger capacitor would be used I guess it will kick off mains fuse because of excessive current pulse. So the only safe way is to slow-down the rising edge of the pulse. I used a big 250W choke from street lighting (with sodium lamps) that smooth the pulse to look like a half of sine wave. The choke is not so small and light but still better than a variac. It didn't fit inside the box so I attach it externally via cables. When there are no filtering capacitors attached to power rail it may cause a risk of high voltage spikes so I connected a varistor in parallel. The choke also limits the peak current flowing through half-brige. When I bypassed the choke MOSFETs blew out within a half of minute of operation. Here is the schematic of power part of the driver:

PSU & half-bridge schematic

and controller part:

half-bridge controller schematic

and some waveforms captured on thyristor pre-regulator without and with filtering:

thyristor regulator output thyristor regulator output with filter cap.
without Cf with Cf

      Further I had investigated the PLL circuit and came to surprising conclusion that I measured output current wrong way all the time. Previously mentioned phase shift was caused by parasitic inductive character of current sensing resistor that I used. I was trying to cancel phase shift that was never existed. So it was not surprising that when I forced additional 90° phase shift into loop it stopped working. Now I measure current via current transformer (that I grab from some old CRT monitor) terminated by 75Ω non-inductive resistor. First I tried this measuring method with resistive load (a bulb and some hi-ohm resistors) to prove it works well. Used CT is not superior but it's good enough to get some rough scope.

current XFMR in circuit current XFMR measuring bulb load current XFMR measuring bulb load-scope

      After connecting TC to the driver it seems that current and voltage waveforms are OK and visible phase shift is quite small. When I manually tuned the PLL via variable resistor (voltage divider) connected to VCOin (without feedback) it proved that the biggest sparks was reached on zero phase shift between output voltage and current (that I seen as 90° before). But nothing is perfect and I can see that phase shift is slightly varying with output power (resp. sparks length):

phase at low power phase at middle power phase at max power
Uout, Iout Uout, Iout Uout, Iout

      So I left variable resistor there connected via 20kΩ resistor to VCOin for phase shift fine tune. I set zero phase shift at maximum output power when it is most important. Maybe when I would use current transformer at the bottom end of secondary coil to pick up feedback signal for PLL instead of antenna it will be more stable but now I didn't get good results with CT.


      I got genius idea how to simply modulate TC output with low frequency signal. I had already seen some "audio-modulated TC" on the Internet but it was made quite a complicated way by modulation of power supply or using PWM modulation of the MOSFET bridge. But PLL circuit is so challenging, it's just needed to couple a source of some audiosignal to VCOin input of the PLL. And you don't need any extra power - a headphone output of common pocket music player is enough for deep modulation of hundreds watts of RF power! This primary cause a frequency or phase modulation but due to the fact that TC is a tuned resonator it will cause also amplitude modulation as a secondary effect. When driving frequency is tuned away from f0 it will cause a significant drop of output power. When the driving frequency is biased e.g. to the left side of resonant curve then positive modulation voltage will cause increase of driving frequency which become closer to f0 and it will increase output RF power and negative modulation voltage will cause decreasing of driving frequency which will be further from f0 and RF power will drop. What is the purpose of this modulation? Such TC can be used as long/medium wave AM transmitter (proper antenna will be needed for longer distance) or as a plasma tweeter. I was surprised how the audio-modulated sparks can play quite well. It is similar to classic tweeter (missing bass) and it can be pretty loud. Signal distortion depends on position of operating point on resonant curve - you have to bias it in the most linear segment. It is done via variable resistor at VCOin that I mentioned before. During operation it can be also heard a hissing noise of the sparks. If we would use high enough driving frequency (somebody says above 3 MHz) then hissing noise will disappear and SNR will increase. Such high frequency can be achieved only with a very small TC (a few centimeters dimensions) and its driving would be very difficult with common semiconductors. There are available some special RF MOSFETs up to 30 MHz from IXYS company but they are not cheap and not available in our country anyway. Have a look at EasternVoltageResearch site, they are developing such small TC called PlasmaSonic project. Here's a cut movie (DivX 5.1; 1,86 MB taken by my digital camera Canon PowerShot A70) of my audio-modulated TC playing Thunderstruck song from AC/DC :). I slightly overmodulated it to make spark modulation more visible. But for small distortion it's needed to keep lower deep of modulation so visible display will seem to be static. When I was trying audio-modulated TC first time I was fascinated by it and spent some hours by playing various music via TC and I also successfully received my AM transmission on radio placed in second room. Here's the final schematic of PLL circuit:

PLL schematic with audiomodulation

RF shielding is important:

PLL board bottom PLL board top PLL board housing

      You can see also a 74HCT04 inverter and jumper on the PCB that allows me to choose inverted or non-inverted signal going from PLL but same effect can be achieved by swapping primary coil terminals so it is not needed. I will mention the effect of inverted and non-inverted driving later. I made a shielding box from 6 single-sided pieces of PCB soldered together. Before I made this shielding there occurred an RF interference that drove my driver crazy. The top side is not hard-soldered but it's hold by 2 wires and can be opened up. Antenna is simply made of 10 cm wire going through a small hole in top side. On the right side is a jack connector for audio input and 3-wire cable (+5 V, GND, TTL signal output) connected to driver base board.

      The process of debugging MOSFET half-bridge was not painless. I blew off about 10 pairs of power MOSFETs. It happened in many cases: when I had detuned and unshielded PLL circuit, when I tried to bypass filter choke in thyristor pre-regulator, when I tried to decrease the number of primary coil turns and during my experiments with primary serial resonant capacitor.
      During my previous experiments I connected TC primary coil without resonant capacitor - cold primary terminal was attached to the middle node of 1µF film capacitor divider between power rail and ground. In this case I got only up to 20 cm long sparks to grounded object while RMS power consumption was around 200 W. When I connected electrolytic filter capacitor to power rail sparks became much shorter but very hot and it was melting the iron spike of discharge electrode. RMS power consumption was increased to around 300 W. Power MOSFETs was only slightly warm so I though that I should decrease the number of primary turns (lower impedance) to increase power and prolong the sparks. I tried it twice but both experiments quickly ended with death of MOSFETs (even I didn't observed any significant warm up) and spark length didn't prolonged. So it seems this is not the right way to go. Here's a photo of some of semiconductors that died during my experiments. BTW I observed that fuse in mains circuit is just for fun because in 90% of cases it blew together with MOSFETs (simply coz it's too slow) sometimes together with clap of mains circuit breaker on the wall. Sometimes it scattered to little pieces and I had to shake out all this garbage from fuse holder.

Silicon Heaven
Silicon Heaven

      Further I mostly used tuned primary coil. On foreign sites this idea was spread too under the name DRSSTC (Double Resonant Solid State Tesla Coil). I would say that my DRSSTC was one of the firsts, constructed in August 2003. So primary coil is in series with a resonant capacitor which tunes primary circuit to the same frequency as the secondary f0. It's good idea to tune primary circuit slightly below secondary f0 because of f0 will drops when sparks appears and become larger. Due to resonance the voltage on primary coil (and capacitor too) will rise to thousands volts and also will transform to higher voltage on secondary side. Resonant capacitor must be robust with low ESR and ESL because of huge current is flowing through it. In my first experiment I used some small disk ceramic capacitors which had been burnt within 10 seconds. I found that high voltage (usually a few nF at 1500 - 2000 V) pulse film capacitors from a CRT TV or monitor deflection circuit works very well. There are available WIMA FKP1 pulse capacitors in GME but they are quite expansive (about 30,- CZK per one) so a serial-parallel network forming a MMC will cost a lot. Currently I use parallel combination of 5,6 + 5,6 + 0,47 nF (11,6 nF together) to make primary resonate at 340 kHz. The effect of scattering resonance into 2 peaks due to high coupling has occurred here. Lower peak is at 290 kHz and higher peak is at 405 kHz. It's interesting that I got much longer fatter and straight sparks at higher 405kHz peak. On lower peak the sparks are more weak, forked and makes little bit cracking sound instead pure 50Hz hum which indicated bad tuning. I can switch between this 2 peaks by including or bypassing an inverter on PLL output or just swapping primary leads on half-bridge output. I don't fully understand this effect yet...
      In this DRSSTC mode I reached 35 - 40 cm long sparks but also the power consumption and power dissipation has rose significantly (up to 800 W without filter caps.). After a few minutes of run time the heatsink temperature reach about 60°C and also the same for resonant capacitor. Even the primary coil wire with 2mm diameter will warm up. But I don't need to run TC on full power for a long time. Maybe I should mount some fan on the heatsink. Here is a table of real power consumption (measured with TrueRMS wattmeter DMM Metex M-3860M, thyristor pre-regulator is fully opened so it equals to half-wave rectifier, 230 V mains voltage) of whole driver in dependence of configuration:

SSTC DRSSTC supply filtering
197W 645W without filter cap., only choke
217W 850W without filter cap., without choke
288W 960W with 220µF filter cap. and choke
495W - with 220µF filter cap., without choke

By using a variac I tried to measure how the spark length depends on total input power:
spark length vs input power

This curve roughly match a square root function. I tried to estimate an empiric function (dashed line) for spark length: l = k*(P-P0)q, [cm; W] with parameters k = 1,7; q = 0,46 and P0 = 110. If a full-bridge would be used instead of a half-bridge, then voltage and current swing will be 2x more and power will be 4x more but sparks will be only rougly 2x longer. Need to say that spark length also depends on other factors like duty cycle, modulation waveform, toroid, etc. So this measurement can be seen only as relative, valid for one specific SSTC.

IGBT half-bridge

IGBT značka 15.8.2004 Now I would like to introduce a new progressive semiconductor switching part - an IGBT transistor. This shortcut stands for Insulated Gate Bipolar Tranzistor. This means it behaves like a common bipolar transistor at output but has an insulated gate so behaves like a MOSFET on input. IGBT is not a hot new part. I have known about it for a long time but I saw mostly the extreme power IGBT bricks which are too slow and so they are useless for driving my SSTC. Recently when I browsed Fairchild semiconductors product list and datasheets I had discovered a few smaller IGBTs with pretty good dynamics parameters which can compete commonly used MOSFETs for building a SSTC. IGBTs combine advantages of both MOSFETs (controlling by electric field instead of pushing huge base current) and bipolar transistors (low saturation Vce voltage - less static power dissipation). When switching higher currents usually the Vdrop on MOSFET caused by its Rdson exceed Vces so IGBT has less power dissipation than MOSFET. And when the junction temperature would rise the Rdson of MOSFET will rise too but the Vces of IGBT will drop. For example let's compare MOSFET IRFP460 (Vdsmax = 500V, Id = 20A, Rdson = 0,27 R) and IGBT FGH20N6S2D (Vcemax = 600 V, Id = 28 A, Vces = 2,5 V @14 A) so static power dissipation of the MOSFET will be 53 W at Id = 14 A (3,78 Vds drop) and power dissipation of the IGBT will be only 35 W. IGBTs are more robust and can handle larger peak current. Some of them are short circuit rated - this means that IGBT can short power supply rails and withstand the short circuit current for some microseconds (time depends on Vge, harder driving - less time). Another big advantage of IGBTs is that its structure doesn't include any body diode between collector and emitter like inside MOSFETs. So then manufacturer can implement inside a hyperfast recovery diode (trr about 35 - 70 ns) or let users to connect any other external diodes. This simplifies bridge structure and also decreases parasitic inductance caused by needed additional wiring around MOSFETs. Gate capacitance of an IGBT is similar or maybe less than Cg of comparable MOSFET. Rated Vge is typically ±20V (±30V pulse). Threshold gate voltage is ussually bigger than for MOSFETs. Dynamics parameters of new IGBTs are quite good. It's interesting that most of IGBTs has significantly (many times) longer toff_delay compared to ton_delay (e.g. 55 / 13 ns). This asymmetry prolongs the cross conduction (while low-side IGBT turns on very fast then high-side IGBT still stays turned on for some time and vice-versa) when a huge current is flowing from supply rail to groud through both IGBTs (if there's not enough deadtime). On the other side short circuit rated IGBTs shouldn't have a problem to withstand this cross conduction current for some tenths of nanosecons but it increase power dissipation. Here's my working table with comparison of important parameters of some IGBTs and MOSFETs which I studied.
      Because The MOSFETs in my DRSSTC driver are very heavily loaded I decided to replace it by IGBT FGH50N6S2D (Vcemax = 600 V, Id = 75 A, Uces = 1,9 V @30 A, Qg = 70 nC @15 V) which easily overpowers used IRFP460 (quite obsolete) MOSFET in all parameters. Gate driver circuit and GDT didn't need to be changed. I just replaced the transistors. DRSSTC started without any problems. It seems to me that IGBTs are slightly less heating. Also sparks was prolonged by few cm (and power consumption has increased - up to 850W without elyt filter capacitors). But at the same time I raplaced resonant capacitor too so it may be rather affected by this. I measured current draw of the whole driver circuit and it was 170 mA with zero voltage over IGBT's power rail and 217 mA under full load at 380 - 400 kHz.
      Just for a test I tried to measure heatsink warm up within 90s on full power. I used digital temperature sensor Dallas DS18B20 placed in a hole in heatsink near 2 cm from IGBT case connected to PC (BTW Dallas 1-wire bus proved a good noise imunity against the radiated field from TC :). Heatsink temperature had risen from 30°C ambient temperature to 45°C (maximum occurred after some delay after shut down due to thermal inertia). So temperature difference is 15°C and I calculated heatsink thermal capacitance to be 388 J /°C. This means that IGBTs must heated 65 W or more. But I don't know how much heat was radiated away so this calculation doesn't give much sense :-\
      Here are some oscillograms of high side MOSFET gate voltage (2 on the left side) and high side IGBT after replacement (2 on the right side). One is taken for minimal power when PLL is not locked yet and other is for max. power.

Ugs on highside MOSFET, min.power Ugs on highside MOSFET, max.power Uge on highside IGBT, min.power Uge on highside IGBT, max.power
Ugs, min.power Ugs, max.power Uge, min.power Uge, max.power

      And finally here are some photos of streamers. The 1st photo on left is without the resonant primary capacitor, next are with Cres. All on full power. And one sparky movie (DivX 5.1; 176 kB). As a further project I plan to build a full-bridge with four FGH60N6S2 IGBTs and with some smarter driving circuit.

SSTC sparks DRSSTC sparks DRSSTC sparks DRSSTC sparks DRSSTC sparks

      8.8.2005 Now I have powerful HV/HF supply so I decided to build a homemade Geisslerovu tube. Very simple and quick experiment can be done with an old glass syringe that we choke by wet finger and pull out the piston so low pressure area is created inside the cylinder. Then we put it in RF field around top of CW running tesla coil. If there's sufficient intensity you will see glowing discharge inside. I don't suggest to try this with a plastic syringe because it will melt and you get RF burns to your fingers from the discharge. It worked for me on the first try but I would like something bigger.
      I chose an elongated glass from olives (height about 14 cm). I drilled a hole through the metal cap and soldered a piece of copper pipe inside. One electrode was made from a needle spike soldered to inner side of the cap. Second electrode is freely lying at the bottom of the glass (capacitive coupling only). It's made of a small screw with grinded spike and washer that stabilizes it in position where spike is directed oposite to top electrode. I connected the copper pipe protruding from the cap to intake of a compressor from an old fridge by a piece of short hose. Output of the compressor was going to another glass with water. I turned on the compressor and I was waiting untill bubbling in water disappear. I have no idea what level of depression was there because I don't have equipment to measure it. Then I strongly pressed the hose by tongs and pulled out one end from compressor intake pipe and stuffed it by a metal roller. There's a risk of sucking outer air inside the glass during this operation. I tried it more times until I was satisfied with the result. When I put the bottom of the glass close to top of running tesla coil the glowing discharge was ignited between two electrodes inside. After 2 weeks the discharge became little bit weak and after some months I was unable to ignite the discharge in full length of the glass. I am aware that it cannot be fully hermetical. This would require some glass-work and solve a sealing of evacuated tube...

primitive Geissler tube Geissler tube Geissler tube Geissler tube

      10.11.2008 I put my DRSSTC half-bridge driver with IGBTs under heavy load on last Teslathon 2008. After longer pulling of arcs to a light bulb a blackout of the room has suddenly occurred. Fuse on the house switchboard was blown but I wonder that small T 6,3 A fuse in the driver survived. I though that IGBTs died but after some later measurement at home I found that only protective varistors on main bridge supply rail was cracked. Diodes in tyristor preregulator died too due to the short circuit. I replaced the cracked varistor pair by three bigger (4,5 kA peak current rated) varistors from Semic.

popped varistors

      When I checked availability of used IGBTs FGH50N6S2D and FGH60N6S2 by Fairchild I found out that this IGBTs became obsolete. A last few pieces can be bought from Farnell and Arrow. So I sent a question to Fairchild tech. support asking for some suitable replacement of these IGBTs and they suggested me HGTG30N60A4, HGTG40N60A4, FGH60N60SF. But I was disappointed by large total gate charge and long turn off delay (see table bealow) compared to older ones. This may be a problem for high frequency switching. They told me that new IGBTs has lower Eon and Eoff energy than older ones. But I guess that in a bridge topology circuit it is much more important the matching of switching times tdon, tr, rdoff, tf to minimize cross-conduction time. I also asked Starmans Electronics distributor and they recommended me IXGH40N60C2D1 alternative by IXYS for a quite good price 149,- CZK inc. VAT. So I just bought 2 IXGH40N60C2D1 for testing and backup. But I would state that old good FGH60N6S2 cannot be overpowered by any of this replacements. Here's a list of important parameters of discused IGBTs:

IGBT type Uce [V] Uces [V] Ic [A] Pd [W] Qg [nC] tdon [ns] tr [ns] tdoff [ns] tf [ns] D-trr [ns]
FGH50N6S2D 600 1,9 75 463 70 13 15 55 50 50
FGH60N6S2 600 1,9 75 625 140 18 15 70 50 -
HGTG30N60A4 600 1,8 75 463 225 25 12 150 38 -
HGTG40N60A4 600 1,7 75 625 350 25 18 145 35 -
FGH40T65SHDF 650 1,5 80 268 68 18 27 64 3 100
FGH40N60SMD 600 1,9 80 349 119 15 22 116 16 36
FGH60N60SF 600 2,3 120 378 198 22 44 144 43 -
IXGH40N60C2D1 600 2,0 75 300 95 18 20 90 32 100
C2M0080120D SiC FET 1200 80 mΩ 31 208 49 12 14 23 18 40
Uces, Ic, Pd, txx @25°C, Uces @ 40A, trr @30 A, Qg @15 V

      In the near future there will be available an interesting alternative: IGBT transistors made of silicon carbide (SiC). They are faster, has lower Vces, doesn't suffer by deep saturation and can work at higher temperatures. It woulde be possible to reduce power losses more than 50% by simple replacement but currently there are only few models available for high prices.
      17.12.2013 UPDATE: good days are comming, prices of SiC semiconductors dropped significantly. I just found a new SiC 1200V power MOSFET C2M0080120D by Cree (see the last line of the table above) that is available on Digikey for 16,67 $ or Mouser for 13,33 euros. This prices are similar to fast silicon IGBTs. This FET offers devil speed with low driving gate power requirement. I guess that running a 1MHz bridge would be easy with it. The only one problem of this SiC FET is assymetric gate drive max. -10 / +25 V. This disable an easy way of direct gate driving via GDT. To fully open the device it requires at least 16 - 20 V gate voltage.

      24.12.2008 I decided that I should make some better discharge terminal for my BigTC and replace that piece of wire by something with a better look. I remember that I saw using of a sparking-plug on some foreign website years ago. So I cleaned the sparking-plug, ground away all sharp edges and polished the metal surface. Then I mounted it on a plexiglass circlet and glued it inside on top of secondary PVC pipe. Sparking-plug body is connected with a short piece of wire to hot end of secondary winding. There's a M4 screw on top of sparking-plug. I screwed on a piece of metal tube with internal threading so I can screw on various metal spikes. When the spike will be burned by strong arcing I can replace it easily. I also tried an ion-motor made as a 4-tail star cut from metal sheet with crooked tails, which I put on the spike of new discharge terminal. During a short time it will rotate at high RPMs. I'm afraid of it could jump off the small dinge and fly away :). New terminal is also more robust so it's easy to put a toroid on it.
      Some time ago I found a small plastic toroid (probably a boat bumper) during a walk. I wrapped it carefully by aluminium foil and put on top of my TC. That toroid with polystyrene body I used before was not so good - it became to collapse due to heat produced by RF currents. This one is made of tough plastic so it should withstand it. Resonant frequency was dropped by 50 kHz. Arc length was not prolonged significantly but sparks became fatter and also it hurts more when touched by hand. IGBTs are not much happy with this and produce more heating so it's not suited for longer run time.

new discharge terminal plastic toroid form DRSSTC sparks with toroid DRSSTC sparks with toroid DRSSTC sparks with toroid

      20.8.2012 A small accident has happened to my SSTC during summer Teslathon 2012. After about 1 minute of happy sparking a white smoke appeared coming out from the power box. I turned it off and blew out the smoke. I saw that it was leaking out of film capacitor Tesla TC207 1 µF / 400 V used in C-divider for power rail blocking and AC coupling the output of half-bridge. Now I know that this TC207 class of capacitors are not good for handling RF and high currents. I replaced it by some more powerful caps 5 µF / 400 V. I also reworked the resonant capacitor to look better. I made new MMC on a piece of plexiglass. I used mostly polypropylene impulse capacitors Tesla TC344 that I formed into 9 parallel strings, each of 3 capacitors (min. 1600 VDC rated) in series (4800 VDC total rated). I tried to slightly vary the capacity of MMC and ended with 11,64 nF (this is very close to old value). The MMC is not heating during operation.

smoking SSTC driver burned bridge DC-blocking capacitor new MMC 11,64nF / 4,8kVDC

      Later during playing with a toroid on secondary I accidentally flip the switch for bypass thyristor pre-regulator that resulted in damage of IGBT pair FGH50N6S2D, R.I.P. they worked for along time and very well, but you know this IGBT = It Goes Bang Too :) I replaced it by my last stocked pair of IGBT FGH60N6S2 without internal diode. I also replaced external diodes by faster Vishay VS-30ETH06FP (600 V / 30 A / 30 ns) and bypassed schottky diodes connected in series with power transistors. They were used to disable MOSFET's internal body diodes but now when I use IGBTs they are no longer needed. Here is updated schematic of power part of the SSTC circuit:

PSU & half-bridge schematic


updated at 3:26; 26.10.2016