Active PFC for Tesla Coils / Tesla Coils / Forums

Forums

4hv.org :: Forums :: Tesla Coils

« Previous topic | Next topic »

Active PFC for Tesla Coils

Move Thread

LAN_403

Steve Ward

Wed Mar 17 2010, 03:08AM

Registered Member #146 Joined: Sun Feb 12 2006, 04:21AM
Location: Austin Tx
Posts: 1055

I figure if i did all this documentation, i might as well share it:

Paper discusses the benefits of active PFC for large solid state tesla coils, and documents my design for a 10+kW power supply. It surpasses the design that is currently on my website which operated in a different mode.

GeordieBoy

Wed Mar 17 2010, 02:04PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Hi Steve,

Top job! A very informative and concise paper. Continuous current mode is definitely the better way to go for this application and power level. But you did a fantastic job getting that other discontinuous boost converter to work so efficiently with the interleaving.

This converter looks very good on paper - How does it perform during testing? Have you done anything like load step tests on the output?

If you are always running the PFC converter in the same country (ie from the same supply) you can do away with the Vrms sensing network. This is only required to implement something called "line voltage feed-forward compensation". It basically makes the control loop work equally well on 110V and 230V supplies so you can sell universal power supplies without requiring a supply-voltage switch on the back that the user can set wrong! If you don't need to run the converter on 110V (which would be extremely hard on the power electronics at full power anyway!) then you can do away with the line-voltage (Vrms) sensing. It improves power factor slightly by getting rid of this because there is always some residual ripple from the line voltage that contributes to the total harmonic distortion of the converter as a whole. (Usually about 1.5% THD is from the voltage feed-forward ripple and 1.5% is from the output voltage feedback ripple. So you can approximately half this THD and improve PF by removing the voltage feed-forward compensation if you don't actually need it.)

Those T650-34 Micrometals cores are monsters. I think that is the best material for this power and frequency, although you could probably even get away with a big Type -26 or Type -52 toroid at 8kHz because the frequency is quite low. Micrometals will probably tell you otherwise though to sell a more expensive material! When you design the boost choke you can also tolerate quite a bit of progressive "saturation". It can be a good thing for the boost inductor to loose some of it's permeability at high load currents, as long as it doesn't saturate hard! 50% loss of perm at the peak of full load current is perfectly normal. It is often desirable because it means that the boost inductance is largest for small load currents, and decreases for high load currents. The increased inductance for light loads helps to maintain the inductor current continuous down to a lower load current, and the decreased inductance at heavy loads improves transient response and current-loop bandwidth when it is needed most.

Your QCW coil is also very impressive. I hope you might find time to reveal a few more of its secrets to us SSTC oldies!

Now that you have a high-power active PFC pre-converter and a high-power synchronous buck converter you have a very flexible programmable power supply for SSTC experimentation. (It's very similar to the power supply and modulator sections of a solid state AM broadcast transmitter.)

Incidentally if you are aiming for high quality audio modulation you can regulate out most of the remaining ripple in the output voltage from the boost-converter by using voltage-feedforward in the buck-converter that follows. This is quite common in the industry: The PFC pre-converter takes care of line voltage variations and implements the power factor correction but only produces a semi-regulated output still with substantial ripple at 2x the line freq. The downstream buck converter then regulates out the remaining 100Hz ripple and tightens up the voltage regulation. This two stage approach is the most common used today because it gives excellent performance. The input and output currents are also continuous "inductor smoothed currents" so it produces relatively little EMI out of either end!

Incidentally, for the QCW coi, did you consider modulating the power throughput by detuning the drive rather than modulating the DC bus supply to the H-bridge? It might be easier as you can apply the modulation as a small signal injected into the oscillator controlling the H-bridge rather than requiring a chunky synchronous buck converter. If you modulate the drive frequency on the high-side of the resonators natural frequency you should be able to control the current draw (and toroid voltage) over a large dynamic range. You may even be able to get a faster transient response because the synchronous buck converter will always be struggling to transfer charge back and forth between it's input reservoir capacitor and the smoothing capacitor at it's output (across the DC bus of the H-bridge.) Just a thought.

Great work, I look forward to reading more about your projects,

-Richie,

Steve Ward

Wed Mar 17 2010, 05:21PM

Registered Member #146 Joined: Sun Feb 12 2006, 04:21AM
Location: Austin Tx
Posts: 1055

Arggg Richie! Your questions get me all excited to reply

I dropped the DCM PFC because i was not satisfied that it would be reliable enough. The magnetics were smaller, but the silicon was much bigger, though likely because it was too overbuilt.

Real testing is great, very stable. I dont remember exact figures of % voltage drop but it was pretty typical for this type of converter (10% of Vmax, perhaps). I often stepped it from no load to ~10kW back to no load with no stability issues with either voltage regulation or current regulation from the mains. When the tesla coils play music, i have to deal with all sorts of peak loading conditions, so i was extra cautious to make sure i got it right!

Another fellow out in California copied this converter design for his bigger tesla coil with very good success, but he replaced the digital control stuff with a pot after his micro-chip processor proved to be troublesome from tesla coil generated EMI. Im not sure why the atmel setup i use was more robust, could be an issue as simple as cabling between the PFC and the TC...

About the Vrms feedforward, i do realize it adds a tiny amount of THD (subject to its filtering), so maybe its worth scrapping it. Would you consider a scenario where the line might be 208V or 240V to be enough difference to warrant the use of the VRMS sensing? I seem to recall at one point testing with the VRMS signal just fed from a DC level when i was trying to improve the line current shape, but i dont recall it proving to be anything particularly better.

As to the core selection, i actually didnt get the cores direct from micrometals, which is why i had to chop 2 cores apart. I think it was just fortunate that the core material was in fact a good choice for this converter. I remember testing this thing at 8kW for like an hour, and the core heating was just miniscule. I didnt realize it was "normal" for a designer to run the core with much saturation, surely i've over-built the inductors by a good 50%! I eventually would like to operate both tesla coils from 1 supply (i have 2 of these PFCs currently for 2 coils) where the peak power draw would be in the 15kW area. I suspect it should hold out just fine as the duty cycle isnt 100%.

Incidentally, for the QCW coi, did you consider modulating the power throughput by detuning the drive rather than modulating the DC bus supply to the H-bridge? It might be easier as you can apply the modulation as a small signal injected into the oscillator controlling the H-bridge rather than requiring a chunky synchronous buck converter.

Your instinct might be better than mine, i dont know, but consider the switching losses of a synch-buck converter at ~40khz vs the switching loss i'd experience with the TC bridge at 350khz. To me it seemed much wiser to maintain my perfect (within 30nS) ZCS on the TC bridge. The primary current on the TC peaks at 140A, the sync buck has to supply about 100A peak (hey, RMS to peak ratio, no coincidence) to the bus.

You may even be able to get a faster transient response because the synchronous buck converter will always be struggling to transfer charge back and forth between it's input reservoir capacitor and the smoothing capacitor at it's output (across the DC bus of the H-bridge.) Just a thought.

Surely you could achieve higher bandwidth this way, but the sync buck i designed had BW to about 5khz, and with feedforward (AKA EQ) you could get it to 10khz i think. For anyone wondering, i did pump "analog" audio through the thing in CW mode with about 15" of plasma on top the coil. The intermodulation distortion inherent to modulating plasma makes it sound nasty, though with just a few inches of spark it was the best audio modulated tesla id heard yet. This leads me to believe that the plasma dynamics are too non-linear at high power and it will never be an awesome loud sound source... until some DSP guru figures out some way to pre-mangle the input signal to compensate.

Anyway, this is pretty off topic now!

GeordieBoy

Wed Mar 17 2010, 08:22PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Arggg Richie! Your questions get me all excited to reply

That's good. It's good to discuss stuff and good to feel proud of what you've built and want to show it off, share ideas, improve it etc.

I dropped the DCM PFC because i was not satisfied that it would be reliable enough. The magnetics were smaller, but the silicon was much bigger, though likely because it was too overbuilt.

CCM is the industry standard for power levels over about 500W and many designs as low as 100W use CCM or boundary-mode DCM for PFC. At high powers switching losses are much lower for CCM because the peak current isn't twice the average line current like it is with DCM. The reduced inductor current ripple of CCM PFC also reduces skin effect and core losses and makes for a whole lot less RF conducted back into the mains supply. PFC designers for commercial equipment that must meet approvals are always struggling to achieve good PFC but keep the boost inductor's current ripple out of the line current! I'm guessing you're not too fussed about EMI back into the supply, but there's a neat technique called ripple-steering that I like to use in high-power CCM PFC boost converters to reduce the RF component of the line current to almost zero.

I often stepped it from no load to ~10kW back to no load with no stability issues with either voltage regulation or current regulation from the mains.

Sounds perfect. A 100% load step is just about the worst thing you can hit it with, so if it survives that without the voltage overshooting then you have no problems.

Another fellow out in California copied this converter design for his bigger tesla coil with very good success, but he replaced the digital control stuff with a pot after his micro-chip processor proved to be troublesome from tesla coil generated EMI. Im not sure why the atmel setup i use was more robust, could be an issue as simple as cabling between the PFC and the TC...

I'd probably keep the microcontroller out of the control loop also - just use it for instrumentation. Although I must say that a current trend in power electronics is to use a low cost DSP to perform all that "current modulator" arithmetic stuff that is done by analogue gilbert cells in typical active PFC chips. A DSP that samples the incoming mains voltage waveform and output voltage level can control a boost converter to give theoretically unity power factor without any distortion by only making discrete adjustments to the current loop's demand level at the zero crossings on the mains supply waveform. That way there is no distortion caused by output ripple voltage modulating the current loops demand signal over each cycle. I wouldn't go to the trouble myself, but it is an area of active research in the Power Electronics community.

Would you consider a scenario where the line might be 208V or 240V to be enough difference to warrant the use of the VRMS sensing? I seem to recall at one point testing with the VRMS signal just fed from a DC level when i was trying to improve the line current shape, but i dont recall it proving to be anything particularly better.

You are right that the improvement in PF isn't particularly profound, i just mentioned it in case you had realised you could loose those components. I'm not sure if that IC has built in under-voltage lockout, so you might want to leave the Vrms sensing in. In severe supply brown-out conditions, active PFC converters can attempt to continue drawing the same level of power from the line which can result in destructively high currents flowing if under-voltage lockout isn't implemented.

If you removed the Vrms feedforward compensation then the supply voltage variation appears as a variable gain term in the transfer function of either the voltage or the current control loop. I forget which. The difference between 240V and 208V isn't particularly large, although I think it ends up getting squared. Whether or not this variation is sufficient to cause problems depends on how stable the control loops are to start with. Obviously the ratio of 270VAC to 90VAC is an extreme case, and when this is squared it ends up making the loop gain of one of the control loops change by a factor of 9:1. It's virtually impossible to achieve acceptable loop compensation under this wide range of loop gain, so that is why the Vrms feedforward compensation was invented. It cancels out the variation in loop gain due to supply voltage so the control loop can operate optimally over a wide range of AC supply voltages.

I didnt realize it was "normal" for a designer to run the core with much saturation, surely i've over-built the inductors by a good 50%!

Micrometals excellent design software will tell you what loss of incremental permeability you are getting with your existing boost inductor design. But if it is working fine and the current control loop is stable then it must be okay. I'd guess that it's already down to about 80% permeability at the peaks of the AC line current under full load, but it might be otherwise. If inductance falls too low due to loss of permeability you usually see the onset of instability in the current control loop near the peaks of the mains supply voltage. This looks like little wiggles in the line current at a frequency near the crossover frequency of the current loop, probably near 800Hz or so. Conversely if the inductor value rises too high with light loads you get something called "cusp distortion" where the inductance is soo large the the line current can't slew fast enough to keep up with the line voltage where it goes through zero. It looks like little bites have been taken out of the load current sinewave at the zero crossings (Class B audio amplifier designers would call it crossover distortion!)

Your instinct might be better than mine, i dont know, but consider the switching losses of a synch-buck converter at ~40khz vs the switching loss i'd experience with the TC bridge at 350khz.

You're right that the TC bridge would be hard-switching to some extent when you detune the drive, but the current level would also fall. A decrease in efficiency might not be too painful because the total power throughput is also reduced when the efficiency is less than optimal. It was just a thought as it might be easier to implement.

For anyone wondering, i did pump "analog" audio through the thing in CW mode with about 15" of plasma on top the coil. The intermodulation distortion inherent to modulating plasma makes it sound nasty

I can't be sure for your design, but when I tried using a Class-D amp to amplitude modulate a CW coil the problem I had was that the RF amplifier (in your case the TC H-bridge) doesn't present a constant impedance back to the lowpass filter that follows the Class-D amp (the buck converter). The corona presented a more or less "constant current sink" load back to the H-bridge which was then reflected back to the output of the Class-D amp. This means that for low output voltages (the troughs of the modulation) the current was disproportionately high and the output filter is effectievely terminated into a low impedance load (the RF amp) and highly damped. Conversely at high output voltages (the peaks of the modulation) the current is almost the same as before! and the filter is now very much under-damped and peaky. I found that this caused a lot of intermodulation distortion because the passband of the filter was actually being modulated by the instantaneous value of the audio modulation!

When the spark is small, or the modulation depth is low, then the filter response stays more or less constant so the audio sounds crystal clear as you found.

This leads me to believe that the plasma dynamics are too non-linear at high power and it will never be an awesome loud sound source... until some DSP guru figures out some way to pre-mangle the input signal to compensate.

You could employ pre-distortion using a DSP or one of those analogue incremental resistance circuits they use to shape one waveform into another in function generators. However I think that it will never be perfect. I personally think the most fruitful way forward for large signal audio modulation of a SSTC would be to FM modulate the drive frequency in such a way that you servo the magnitude of the TC secondary's current to match the instantaneous audio waveform's value. In other words you make a feedback loop where you continuously control the inverter drive frequency (a VCO or DDS) so that the secondary base current (and hence the TC discharge current) is always proportional to the incomming audio signal. Since there is no Class-D (or buck converter) reconstruction filter required the response of the whole control loop should be very fast, and enough negative feedback around it should automatically linearise the horribly non-linear load that the TC corona reflects back to the driver. (What i propose is similar to how big AM broadcast transmitters sample the antenna current and feed that back to the modulator to improve the overall linearity from the microphone to the airwaves.)

Anyway, this is pretty off topic now!

I wouldn't worry. I'm sure there are lots of people reading what you have written with much interest, just not chiming in yet with their own ideas, or making notes for their own future projects. I, for one, find your work very interesting. I just wish I still had some time and space to work on Tesla Coils myself!

All the best,

-Richie,

PS. Do you have any scope pictures showing the envelope of the E-field around the coil when it was generating the ramped bursts and making the long sword sparks? I would be very interested to see a scope picture showing the ramp reference signal generated by the microcontroller, alongside maybe the secondary base current and the HV E-field in the air around the coil. I'll bet that the E-field tops out and levels off fairly early in the progression of the ramp. It would also be interesting to see what happens to the base current and E-field envelopes when you run your ramp with the boost at the end to produce the sparks with florets at the ends! Those would be some very interesting experimental results!

TheBoozer

Thu Mar 18 2010, 01:20PM

Registered Member #1535 Joined: Wed Jun 11 2008, 11:37PM
Location: Northeastern Pennsylvania - USA
Posts: 117

Steve: Thanks for providing such a clear and concise document detailing active PFC.

In regards to a question thrown at me when I was off guard, many modern industrial welding machines utilize Active PFC. Now that this has inspired my curiosity, I've begun reverse engineering the designs. The ones I'm focusing on now have a UC3854N as the heart of the controller.

Must be a common IC, I found the datasheet at DigiKey.

Rich

lightlinked

Sat Mar 20 2010, 01:14AM

Registered Member #2087 Joined: Tue Apr 21 2009, 08:32AM
Location:
Posts: 115

can you give the output of the PFC directly to your DR/SSTC or do you still need another regulator/converter after the PFC? i'm sure many of us are interested in abolishing the variac in favor of a flexible DC power supply

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob