3rd September 2021
Just recently I finished my PLL driven SSTC designed for my university, and even though it was a struggle, this project gave me a lot of new skills and ideas. I really like the idea of clean audio played with the arcs, the only problem is that the coil has to run in continuous mode. My first PLL SSTC is capable of that, but it overheats after like 10 minutes and it is just a small coil with a modest power draw of 1kW. I started new project, a second PLL SSTC which would be able to run continuously without overheating for a long time. It should also draw as much power as possible from 16A mains outlet.
The overheating problem
The biggest issue when it comes to overheating are not the switching transistors, it's the secondary and primary coils themselves. My SSTC I is switching roughly 1kW and the transistors reach a temperature of 60C after about 10 minutes, but they are placed on pretty small heatsink with only little air circulation. To be fully prepared for multi-kW power draw I decided to go with double full bridge of IRFP460 transistors placed on huge heatsink with powerful fans blowing a lot of air. But then again, I'm not that worried about the transistors, the real concern are the coils. For primary coil I bought 25mm^2 (5.5mm diameter) cable for total of 8 turns. The secondary coil is a bit experimental. Instead of lots and lots of turns with thin wire, I have to go the other way to evade overheating. So I decided to use 0.5mm wire and wound ~820 turns on 12.5cm diameter PVC pipe. This way the secondary should not overheat so easily and even though it has only 820 turns, thanks to the much lower resistance of the thick wire, the Q factor here is almost 270. Tesla coils usually have secondary Q in the range of 100 to 200. My 270 Q factor should counteract the lower voltage induced due to lower winding count.
Winding of the secondary
Finished secondary with 3D printed base
Finished secondary with topload and primary coil
Measuring the resonant frequency
15th September 2021
I have used double coated wire, sprayed with insulation varnish and covered with kapton tape. This should give me plenty of insulation. Topload was bough from eBay, it is 25x6 cm. I couldn't use my classic style 3D printed and aluminum tape covered topload, in CW mode this is no longer a solution as it overheats very quickly. With this configuration I achieved resonant frequency of 314kHz and the secondary only measures 28ohms of DC resistance! That's 10x less than most coils of this size, but then again it only has 820 windings.
The power supply for the primary coil is a double full bridge of IRFP460 transistors placed on a huge heatsink with two 5W fans blowing air to cool it down. Each pair of IRFP460 also has its own antiparallel diode DSEI60-06 in TO247 package. There are therefore total of 12 TO-247 package devices.
Each transistor pair also has an RC snubber network consisting of 15R 8W resistor and 330pF WIMA capacitor. This RC snubber acts as a low pass filter to suppress any high frequency voltage transients.
These voltage spikes are caused by parasitic inductances and capacitances which oscillate together. On these oscilloscope captures the voltage on the bridge is roughly 60V while the peak of the voltage transients is about 100V. Supposed that these voltage spikes are proportional to the bridge voltage, we would have 540V peaks when powered from mains voltage! IRFP460 transistors used in this bridge are only rated 500V which means we have to deal with these spikes. There are a lot of ways to handle them, I choose to use RC snubbers to filter them. I originally used 1nF and 15R combination, this RC snubber has a cutoff frequency of ~10MHz (We can calculate low pass filter cut-off frequency as 1/(2*pi*R*C) ). Since the frequency of the voltage transients was measured to be ~34MHz, this RC snubber combination should work just fine. Unfortunately the RC snubber got hot really fast. At 120V at the bridge, the solder on the resistors was molten! This combination wastes too much power that even 2W resistor is literally melting. Calculating the power loss on an RC snubber is not as easy as it might seem. My naive approach to calculate it was to calculate the current which will be flowing through the capacitor and then determine the power lost on the resistor as P = R*I^2. We can calculate the current because we know the frequency at which our bridge is switching and we also know the capacitance and the voltage. This way we can determine the impedance of the capacitor which will be limiting the current (supposed R << Xc). My bridge is switching at 315kHz, capacitance of the RC snubber is 1nF and the voltage on the bridge will be the mains voltage (325V). Therefore we can calculate the impedance as (1/2*pi*f*C) which in this case results in 505ohms. Maximum current then should be 325V/505ohms = 0.64A. Now we can determine the power loss on the resistor as:
P = 15R*(0.64A)^2 = 6.1W
Well no wonder that 2W resistor will overheat quickly at full mains voltage but let's now calculate the power loss at 120V:
I = 120V/505ohms = 0.23A
P = 15R*(0.23A)^2 = 0.79W
Only 0.79W was already melting my 2W resistor, that sounds weird doesn't it? Well as I already stated before, this approach to calculate the power loss is quite naive. Since this calculation only takes into consideration the 315kHz signal, but it doesn't take into consideration the rest of the frequencies that are superimposed on the signal. For example those switching transients have frequency of ~34MHz and at this frequency the impedance of the 1nF capacitor is only 4ohms!
Instead of hardcore calculations I decided to use brute-force. I used 4 2W resistors for total of 15R at 8W. I thought that the problem was solved, but at higher voltages at the bridge (>250V) the resistors started to get really hot again. The current is just too high, it dissipates much more than 8 Watts. So what can we do? We can decrease the resistance, this will lower the power loss, but it has disadvantages. Lowering resistance means that our cut-off frequency will get higher, maybe so high that the voltage transients will be unaffected. Also at higher frequency the assumption that R << Xc doesn't necessarily hold anymore and the resistor is the main current limiting factor. If we make our resistor too low, we risk that the currents flowing through the transistors will be too high. Better solution will be keeping the resistor, but lowering the capacitance. By lowering the capacitance we also increase the cut-off frequency, but that's a disadvantage we can't avoid anyhow. The only capacitors with less capacitance I had laying around were 330pF. With this configuration the RC snubber's cut-off frequency is 32.15MHz, which is still just a bit less than the frequency of the voltage transients.
Drain-Source voltage (yellow 20V/div) and Gate-Source voltage (red 5V/div)
On this scope capture you can see that the voltage transients almost do not exist anymore.
This is the overall schematics of the double full-bridge:
Double full bridge schematics
16th September 2021
Gate Driving Transformer (GDT)
This time I have calculated everything necessary for optimal GDT design. I have chosen a material with inductance factor (Al) 91nH. I do not know which exact material it is but it's exactly this ferrite ring bought from TME.eu. 91nH is quite low inductance factor, but to transfer a lot of power, we need small inductance. In some designs GDT can get away with materials with much higher inductance factor, but when we are driving quite huge double full bridge then smaller inductance factor is desirable.
I have written this handy little script in Matlab to tell me everything I need to know about the GDT.
First we need to decide what is the maximum acceptable magnetic induction so that the core will not saturate. For materials like iron, this is usually quite high, in orders of few Tesla, but this core is ferrite and has much lower magnetic induction capability. I would not recommend to go much higher than 100mT otherwise your GDT can saturate which can cause destruction of both the bridge and the driver! To be on the safe side I decided to go with 50mT maximum magnetic induction. After filling in all the other parameters, the script tells me how many windings I need on the core without the magnetic induction inside the core reaching higher than 50mT. It also tells me the primary inductance and no-load current flowing through. With more than 2A of no-load current, I really have no concern about the transformer not being able to drive my bridge.
It is optional but it's good to know what is going to be the resonant frequency of the secondaries with the Gate-Source capacitance of the transistors. Single IRFP460 has a G-S capacitance of around 4nF, therefore I fill in 8nF because each secondary will drive a pair of these transistors. This resonant frequency then should be way lower/higher than the resonant frequency of the coil.
You can download the script here:
To be on the safe side, I used 10 windings. This way I'm even further from magnetic saturation.
Even after thorough calculations and thought-through GDT design I was left with this disappointing waveform:
Gate-Source waveform 10V/div, 500ns/div
Gate driving transformer
As an experiment I wound another GDT with the same core size but with different material. I used a material with Al = 2900nH. This should drastically change the waveform but it didn't. I was left with the same waveform, which kind of confused me. I have tried different windings count; 5, 10, 15 but nothing helped. It took me whole day to realize that this is caused by leakage inductance of the transformer. I wound the same core as before but with enameled wire with much thinner insulation.
This solved the issue and I was left with this beautiful square-wave:
Gate-Source waveform 10V/div, 500ns/div
Phase Lock Loop (PLL) driver
I'm using the same driver as I did with my first PLL SSTC but this time I designed a PCB for it and used more robust gate-drivers.
Thanks to SMD this driver is only 8x6 cm. It also fits well into a handy little box I ordered from Mouser.com. This box acts as a heatsink and as a shielding from EMI. Overall it took me few days to design and around an hour to solder. The driver is pretty much plug and play, the only thing needed to adjust before turning it on is the desirable operating frequency for which I added a few trimmers.
Here is the overall schematics:
The idea behind the driver is quite straight-forward. It uses a current transformer to sense feedback from the secondary (it could also be connected to sense primary current for DRSSTC), this feedback is then shaped to a square wave thanks to U1 (74HC14). U7 (CD4046BE) chip is driving the transistors at the frequency which can be set with RV1, RV2, RV3 and C21,C22,C23 capacitors. 4046 will generate this frequency even without feedback, but with feedback implemented, 4046 will lock on the feedback frequency (if the frequencies are similar). This is very reliable way of driving the transistors, in case of feedback malfunction the coil will still be driven at a frequency at which the transistors will survive.
Because 4046 generates the driving signal with voltage controlled oscillator (VCO), we can use clean audio modulation. We can let the 4046 chip to lock on the secondary resonant frequency and then modulate the oscillator frequency with analog audio signal. This way we actually force the chip to oscillate at frequencies slightly below or above the secondary resonant frequency which drastically changes the output power thus generating audio as the size of the sparks changes with power output.
For gate drivers I used IXDI/N_630 FET drivers capable of 30A of peak output current. They also have internal under voltage lock out circuit (UVLO) set to 12.5V.
They also have Enable pin which is always a nice thing to have.
I use an NTC at the input to protect the linear voltage regulators from high peak currents as they are quite sensitive to higher current spikes. With only a DC-blocking capacitor in series, you can feed audio directly from your phone to the pin 9 of 4046, but you should always use some kind of galvanic insulation. I solved this issue with a Bluetooth module which generates the audio signal being transmitted to it.
I originally used 100nF as C17 capacitor, but I found out that 1uF handles audio with a lot of bass better.
I will release all the files for the driver when it's ready, but now as I'm writing this article, there is still a lot of things about the driver that need to be improved.
First light (>2.5kVA)
Finally after everything was put together the coil was ready for first test run. I added a switch which selects between two power modes. First mode is single diode rectification with no smoothing capacitors. In this mode, the coil draws the least power but makes the longest arcs. The second mode is full wave rectification with 7800uF of smoothing capacitors. In this mode the arcs aren't that long but they are very thick and hot. In this mode we can also play clean audio with the arcs.
Single rectification mode, 40cm sparks, 1.5kVA
Touching the output in 50Hz mode
Playing audio in CW mode, 25cm arc, 2.5kVA
CW mode, 200V input, 2.5kVA
CW mode, 120V input, 1.8kVA
Copper electrode liquified in CW mode (temperature > 1100C)
New electrode made from graphite rod taken out of 9V batery