Before I could get on with building the PFC into an enclosure, I had one last design decision to make: What sort of EMI filtering to use. The size and shape of the EMI filters would affect the rest of the mechanical design. Ok, that’s management speak for “How am I going to get all this cr@p into the 3U rack enclosure I’ve already purchased?”
Now, I deal with EMC in my day job and am vaguely familiar with the standards and test procedures, but this is a one-off handmade power supply for a Tesla coil. It’s never going to get tested for emissions, and the emissions from the Tesla coil will dwarf the contribution from the power supply anyway.
So the main purpose of the EMI filters is to protect the PFC from malfunction or damage caused by the Tesla coil emissions. These tend to be common mode transients caused by ground strikes, containing frequencies up to the 10s of MHz. There isn’t a great deal of VHF or UHF energy due to the length of the spark channel. So they really aren’t super hard to filter out.
I prefer to connect the filters so the Y capacitors (jargon term for the capacitance between lines and ground) are at the end connected to the outside world. My reasoning is that I’d rather any incoming transients were dumped to chassis ground through the capacitors, than potentially flashing over a choke.
I started by trawling the RS, Farnell and Mouser catalogues for ready-made EMI filters. I ended up with a Delta 30TDVST2 for the input and a Schaffner FN2200-25-33 for the output. These both had the Y capacitors at the load end, so would have to be used backwards from the maker’s recommendation.
I soon discovered a serious problem with both filters: a very high Y capacitance. This isn’t a problem in the intended industrial application, but a bit of a show-stopper for mine. When the PFC is used on a single phase supply, the high capacitance causes enough earth leakage current to trip any RCD. Note that the Y capacitance of the DC output filter also contributes to the leakage, because the DC output is not isolated from the mains and has AC superimposed on it.
I couldn’t find any better filters, so I broke them open and set about reducing the Y capacitance.
I lifted the connection between Y capacitors and earth, and added a 68nF capacitor in series, with a 2.2M discharge resistor. This should give a total leakage current budget of around 10mA at 240V AC. (Odin has 2 68nF capacitors from DC bus to ground already, which contribute too)
The modified filters no longer tripped my house RCDs, so the job was done. In hindsight, I wouldn’t buy ready-made filters again. It would have been cheaper to buy the parts and make them.
The PFC engine is working, but there are a few other things needed to make it usable. (Operationalise it? Or Heaven forbid, weaponise it? 🙂 )
Auxiliary power supply: The PFC needs a small amount of power to run its own control electronics. I decided to use a Meanwell WDR-120-24 switching power supply to provide 24V DC. This is an industrial grade unit that will accept any input voltage from 200 to 500V AC.
The WDR-120-24 is a bit more expensive than the usual 85-265V input range units, but vital for my goal of being able to run the PFC off either 230V single phase or 400V 3 phase power, without any kind of voltage selector switch that could cause carnage if set wrongly.
Precharge: The bus capacitance of the DRSSTC is very substantial. Odin has 4700uF after a recent upgrade. The PFC itself will also need another 1000uF to allow it to work without the DRSSTC connected. All of this has to be charged to the peak value of the mains voltage before the PFC can even start, in an orderly manner without tripping any breakers.
I chose a capacitive ballast for this job, consisting of 22uF motor run capacitors with 10 ohm resistors in series. The capacitors do most of the current limiting while the resistors protect the capacitors and main contactor from the surge when the capacitors are shorted out. The resistors are attached to the main heatsink and protected by the overtemperature cutout.
The precharge controller is based around a time delay and voltage sensing relay. (Schematic in a future post) The voltage between D1 and D2 must get over 200V, and the voltage between D2 and U2 below about 20V, before the sensing relay will pull in. This energises the main contactor, connecting the PFC input rectifier directly to the mains, and powering up the PFC controller through its auxiliary contact. The PFC then goes through its own soft start procedure, charging the DC bus capacitance to full voltage.
Dump load: The large DC bus capacitance also needs discharged when the system is powered down. My previous coils all relied on bleed resistors and took over a minute to discharge. For this build I decided to try some PTC thermistors from Epcos. (Details in a future post.)
The main advantage of PTCs is that, unlike normal resistors, they limit their own temperature and won’t catch fire or explode if the switch controlling them accidentally turns on while the DC bus is powered. This allows me to switch them with a SCR which was already present in the bypass diode module.
EMI filtering: This is as much to protect the PFC from damage by the huge transients generated by the Tesla coil, as to protect the mains from the hash thrown out by the PFC. My search for suitable off-the-shelf EMI filters is documented in another post.
The low power test was a success, so a larger dummy load was obtained- 3x 240V, 3kW water heaters connected in series.
The PFC was placed on a special tea tray test fixture and temporarily wired into the electric cooker supply in the kitchen.
According to my smart meter the result was 8.1kW input, though the other house appliances were using a few hundred watts at this point. The line current waveform looked nice. I used an EMI filter this time so the garbage at the switching frequency is reduced.
The moment of truth could be put off no longer 🙂 The PFC was tested at reduced input voltage using a 110V, 100 watt heater as a load.
In the previous post I mentioned that I couldn’t drive the MOSFETs directly from the controller chip because they need a negative voltage to turn off. The easiest way to provide this was with an isolated gate driver using the excellent Silicon Labs Si8271 IC.
Due to the tiny size of this IC and the requirement for a very tight layout to minimise stray inductance, I designed a PCB for it. This was also an excuse to finally start learning Kicad after years of using Eagle.
Power is supplied by a Murata MGJ6 DC-DC converter connected to J2. +15V to pin 1 and -5V to pin 2. The same DC-DC converter powers the opamp on the voltage feedback isolator board.
R7 is a small value like 22 ohms. I installed a 10k resistor in place of C6 to pull down the input if it became disconnected. R2 was not fitted and R1 was a 0 ohm link, the opposite of the schematic (I got the enable pin logic the wrong way round)
I uploaded the Gerber files to JLC– a Chinese manufacturer based in Shenzhen who ship worldwide- and had the boards the following week for a cost of about £5. JLC are 1/10 the cost of any UK PCB house and have really lowered the barrier to entry for hobbyists.
All this power is nothing without control.
Luckily by using silicon carbide power devices, we were able to handle all the power we needed with a simple boost converter switching at 100kHz. No need for multi-phase or fancy energy recovery snubbers- it’s just a domestic appliance-sized PFC front end scaled up.
This means we can also use a controller intended for said domestic appliance PFC. I used the UCC28180 PFC controller chip from TI, because it only has 8 pins and is marketed as “easy to use”, therefore it can’t possibly go wrong! 😀
(OK, I bought the evaluation board and tested it with some Tesla coil-like loads before committing)
This is the circuit I ended up using. It’s quite different to the original TI application circuit, so I’ll describe the differences.
Compensation: TI provide 2 full pages of formulae for calculating the compensation components, and ain’t nobody got time for that! So I took the values from the evaluation circuit, and doubled the capacitors and halved the resistor because it seemed like a good idea.
Gate drive: Silicon carbide MOSFETs need a negative gate voltage to turn off, and the UCC28180 doesn’t provide this, so I had to make a separate gate driver, which I’ll describe in the next post. The Silabs isolated gate driver chip needed a 5V supply and a 5V signal, so I had to add a level shifter and a 5V regulator.
Inductor current feedback: The UCC28180 is designed to take feedback from a resistor in the DC bus negative. This is a very common design choice in small PFCs. I added the 4x 1N5819 diodes and associated circuitry to adapt the signal from the CTs to look like it came from the original resistor. The 13V zener is used to clamp the spikes when the CT cores reset. A value of 13V guarantees reset even with the maximum duty cycle allowed by the controller chip.
I tested the CT feedback on the UCC28180 eval board before committing to building the full sized version, in case the chip turned out to be unhappy with it for some unforeseen reason, but it performed identically to the original resistor feedback, as far as I could tell.
Voltage feedback: Since the CTs provide isolation of the current signal, and I ended up using an isolated gate driver, I decided to isolate the voltage feedback too so the control circuit would be completely isolated from the high voltage side. I used a simple circuit based on the HCNR201 linear optocoupler and a TL071 opamp.
The voltage feedback input of the UCC28180 has a high enough impedance that it can accept the current from the photodiode (4.5uA per 100V DC bus voltage) directly with no amplification or buffering. This simplifies the circuit considerably and provides some nice synergy. The UCC28180 has circuitry to detect an open feedback network, a dangerous condition that can cause the output voltage to run away, blowing up expensive pieces of power electronics. By configuring the voltage feedback in this way, the protection still works for an open circuit anywhere in the signal path, including forgetting to plug the isolator into the main board, and loss of power to the isolator’s opamp.
Both circuits were easily constructed on protoboard with through hole components (and an adaptor for the UCC28180)
When I started designing I already knew that the heart of the thing would be silicon carbide MOSFETs and diodes. I heard so much from Anders Mikkelsen of Advantics raving about how awesome they are. 🙂
On recommendation from Anders I chose the C3M0065100K MOSFET and C4D10120D diode. These are available from Farnell, Mouser etc. for about £10 each. I figured that 2 in parallel would allow me to run an inductor current of 32A RMS, a reasonable match to the mains supplies available in the UK and Europe. Anders also kindly donated an inductor capable of handling 32A RMS. 🙂
The remaining design decisions were set by the dumpster find of a large heatsink and set of 3 Semikron dual diode bricks. Having such a beefy oversized rectifier made short circuit protection easy: I just needed to come up with one more heavy duty diode to bypass the delicate SiC diodes, and then a short on the output would simply pop the circuit breakers on the mains input.
CT1 and CT2 are small ferrite cored 1:200 current transformers sensing the MOSFET and diode currents. Their outputs are combined in the controller to give the inductor current.
SiC MOSFETs can switch incredibly fast, so a good layout with minimal stray inductance is important. I decided to make a multi-layer structure out of pieces of single-sided copper-clad PCB and copper sheet.
I didn’t manage to fit as many capacitors as I wanted: only 5 on the input and 4 on the output. The output ones in particular will have a hard time with serious amounts of HF ripple current. They are only there to filter out the HF ripple. Large electrolytic capacitors will be needed to handle the power frequency ripple, they will be mounted somewhere else and connected by wires.
Clamps for the MOSFETs and diodes made from pieces of Ikea cabinet legs. 🙂
I decided it was time to build a new power supply for Odin with power factor correction. (See all posts on the PFC)
Power factor what? To explain, we have to go back to the turn of the last century and the War of the Currents. Alternating current won so all of our mains supplies are AC. However, solid-state Tesla coils, like most power electronics, run off DC. Odin needs several kilowatts of power at about 750V DC.
The simplest way to achieve this is to use a voltage doubling rectifier on the 240V AC mains. I used this in my previous large solid-state Tesla coil, the OLTC2, with one small refinement. I used SCRs instead of diodes to allow soft starting and varying the output voltage.
The major drawback of this circuit is a very poor power factor, especially at reduced output voltage. It is particularly bad in a DRSSTC, as it has a huge capacitor on the DC bus for energy storage, and the rectifier charges this directly. I improved the power factor somewhat by connecting a large iron cored inductor in series with the incoming mains. This worked well enough, and you can see details of the old power supply here here and here.
I used this same power supply topology in Mjollnir, Little Cook and Odin. There was no room inside the smaller coils for the passive PFC inductor, so I used a much smaller one, making the power factor even worse. The shortcomings really became apparent in a few shows I did with Odin. It performed really badly when running off a generator, and there were two venues with 3 phase supply that I couldn’t use.
After much thought I decided to go for the simplest possible solution: a boost PFC. These are very common in larger consumer electronics, and controller ICs and design information are readily available. It is also easy to modify for 3 phase input. Essentially you just replace the rectifier with a 3 phase one. The line currents aren’t nice sine waves any more, but the power factor is still greatly improved over a plain rectifier with the same filter capacitance.
Probably the biggest drawback for Tesla coil use is that the output voltage can only be greater than the peak value of the rectified input voltage. You are limited to a minimum of about 400V when running off 240V single phase, and 600V from a 415V 3 phase supply. This is a problem for me at least because I like to check the tuning at reduced voltage after setting the coil up at a new venue.
Eventually I persuaded myself that I could live with this and set to work designing a boost PFC.