High Voltage Forum
Tesla coils => Solid State Tesla Coils (SSTC) => Topic started by: Laci on August 08, 2018, 10:08:48 PM
-
Hello everybody!
Recently I got myself some IRG4PC60U IGBTs and made a fullbridge of them,which will hopefully power my new coil.The problem is the same as always before when using IGBTs:slow rise and fall time.I'm not sure if the waveforms are good or not,but a short try on the actual coil at 20V input was not that good.
Here is a picture of the GDT type TX42/26/13-3E25 with 11 turns on it:
(https://i.imgur.com/ta6UzgQ.jpg)
...and the driver board:
(https://i.imgur.com/XYFAAbk.jpg)
(https://i.imgur.com/feQfbb0.jpg)
I have a 1000uF electrolytic in combination with 1uF WIMA MKS-4 in parallel with a 10uF tantalum capacitor at the both driving MOSFET pairs.The GDT is connected experimentally with resistors and capacitors.
Here are the waveforms:
(https://i.imgur.com/kAprUKI.png)
(https://i.imgur.com/pu9nVF2.png)
Is there any way to make the rise and fall times faster to be able to run the driver up to 170kHz?
-
Apart from making the right selection for the IGBT's/Mosfets I learned that driving them the best way possible often is key to success.
In most cases that mean using a drive circuit that allows for the right voltage and high current with good peak performance.
A GDT for this purpose is best done on the right core for the frequency rage, it should also have suiable saturation characteristics.
Same for the wire used, ethernet cable with insulation is a bad choice IMHO.
Quiet thick magnet wire is a better choice, litz wire even better.
The turns should be tight and as even as possible, if in doubt use toothpicks as spacers on the outside.
But in todays times I really think a dedicated and fast Gate Drive IC would help you most.
With a suffiently strong power supply they not only provide fast rise and fall times but also offer a lot in terms of protection.
Some offer half or even full bridge support in one chip, other can be used in a positive-negative combo.
Either way they ensure no mosfet or IGBT can latch up or turn on together with the wrong partner.
You also eliminate a possible cause where interference can enter the system as you won't have the long and additional wires from the GDT.
For GDT's I won't use any cores from line filters, usually they are quite big and come in white or yellow.
From my experience they are of no use for the high frequencies we use.
Cores from 60 or 120kHz switch mode supplies work reasonably well and I used them for up to 200kHz with no problems.
Size matters to if you ask me and I prefer to have the core just big enough to hold a single layer of the required turns for all wires.
My prefered option is to use really thin wire, like from a cheap syncronous motor (less than hair thin) and to make several lengths of litz wire from it.
First making long "ropes" from just 3 wires.
"Fold" in half and twist again to get a rope with 6 strands.
And then the same for another two rounds to end up with 24 strands.
This is enough for most of my applications but since it is quite quick I sometimes go to 48 strands or even 92 for things that require more power.
All required strands are then fed through the core at the same time so they always stay parallel to each other and won't twist.
Spacing on the outside should be so thaton the inside all strands touch each other, should look like thick spokes on a wheel.
What is used for the connection should be twisted together right from the core and until the connection points are reached.
Of course the distances should be kept as short as possible.
In case you want to try with simple wire first salvage some line filter cores for their wire or better still use the (copper) wire from the primary of a MOT.
Clamp one end in a vice or twist it around something sturdy and with a tight grip run an old rag along the wire - this will remove any leftovers from the varnish so it won't act like needles when you wind it.
Use a pair of pliers or wrap the wire around a grip to slightly pull on it until perfectly straight.
Cut to the estimated lenght you need and wind it with care to prevent kinks.
Again wind all strands at once and in parallel to have the inner surface of the core covered while leaving small gaps on the outside.
Another way out is to use mosfets to provide the gate signals.
For fast switching og IGBT's you need high currents and a fast enough driver, mosfets can do this quite good.
-
Have you tried changing the value of the gate resistors yet?
I find that it solves the problem most if the time
When in doubt lessen the gate resistor value
Also, 40ns seems to me to be quite the reasonable rise/fall time.
The waveform is great otherwise
-
I agree with TDAF. Try changing the gate resistors to a lower value. You'll get slightly higher overshoot but it will improve rise/fall times.
-
I do not think you can increase performance by much as you are already switching faster than the datasheet for IRG4PC60U specifies. 42ns Trise time, Rise time should be measured from 10% to 90% of the rising edge! Not including the overshoot!
You can drive them faster than the specified hard switching maximum, if you drive them in a resonant circuit, so this is your largest gain of performance.
You are driving them with a TC4429 which is a 6A peak driver, do you have any more powerful drivers like the UCC21xxx series that can do 9A?
What gate resistor value are you using?
Switching waveform looks really good by the way, and as others noted, your overshoot is not too large so you can try a lower gate resistor value, if the driver is not already maxed out. The slight kink in the rising edge is the miller plateau :)
Remember, the optimal way of driving a bridge as of follows "To switch the current as fast as slowly possible", what this means is that going really slow you have too high losses, but switching too fast and you have other kinds of losses, you have to find the middle way and I think you are exactly there.
-
@Downunder35m
I know my GDT is not the best for this purpose,it has many leakage inductance and resistance caused by the long wires,but tried the same Cat5e cable in the past with the main insulation removed-no significant improvement.Also tried many half/fullbridge bootstrap IC's from the IR21xx series and they are not compatible with the requied 24V driving voltage.I'm usually following other working designs,so there must be a problem,not sure what...
@TDAF and @oneKone
The gate resistos are 4.8 ohm.It is the lowest value I have and never seen lower than that using by others yet.
@Mads Barnkob
The posted waveforms were captured without input voltage applied to the bridge.This way I should be able to run them to their maximum possible frequency,right?!
Those TC4429 chips are instead of the single two channel UCC driver from the UD1.3.They power the output MOSFETs on the four same heatsinks.
The 24V power supply is a computer ATX PSU from -12V to +12V,which is probably a bit weak at the -12V side,but tried with a 20V 2A laptop PSU and no difference.
Last time I was using a fullbridge of HGTG30N60A4s and I do not remember the waveforms,altought they died at the first mains rectified test,they were working as poor as the new ones now.For some reason after two of the 30n60s died,the remaining two were working for a seriously long time of more than a month!There might be a problem with the fullbridge configuration.
-
@Mads Barnkob
The posted waveforms were captured without input voltage applied to the bridge.This way I should be able to run them to their maximum possible frequency,right?!
Those TC4429 chips are instead of the single two channel UCC driver from the UD1.3.They power the output MOSFETs on the four same heatsinks.
The 24V power supply is a computer ATX PSU from -12V to +12V,which is probably a bit weak at the -12V side,but tried with a 20V 2A laptop PSU and no difference.
Last time I was using a fullbridge of HGTG30N60A4s and I do not remember the waveforms,altought they died at the first mains rectified test,they were working as poor as the new ones now.For some reason after two of the 30n60s died,the remaining two were working for a seriously long time of more than a month!There might be a problem with the fullbridge configuration.
You are around 30 ns rise time, you should not worry about this as you are already below datasheet standard. You will be able to switch faster once you have some resonant current going through the bridge as this will speed up the whole depletion/hole injection process of the FET.
You are go for launch, do not change a thing :)
-
I think there is some confusion here - the rise-time is NOT 40ns (the number above the cursors on the scope) - 40ns is the horizontal offset from the trigger point. The cursor delta-t is shown on the right and is a LOT higher (590ns, maybe closer to 400ns for a proper 10%-90% risetime measurement).
So I agree that it seems a little slow. The actual waveform is very nice though.
Maybe it would have to have a proper schematic of the driver board? The construction of the GDT itself looks pretty nice (though I have not looked up the material specifications of the ferrite). Also note that a) the IGBTs have a reasonably high gate charge for TO-247 devices and b) don't seem to have a co-pack reverse diode (required for DRSSTCs) - are you supplying an external one?
-
The -12V from an ATX is not just weak but also not really suited to join the +12V rail.
The first if for data purposes, the second for high power use.
I would not be surprised to see the voltage break in quite a bit.
No matter the supply I always found it beneficial to have a beefy electrolytic cap close to where the gate drive draws the most power.
And some switch mode power supplies really hate going from idle to full load.
But I agree, some schematics could help to figure out where the culprit is in this nice design....
-
For GDT's I won't use any cores from line filters, usually they are quite big and come in white or yellow.
From my experience they are of no use for the high frequencies we use.
Cores from 60 or 120kHz switch mode supplies work reasonably well and I used them for up to 200kHz with no problems.
Size matters to if you ask me and I prefer to have the core just big enough to hold a single layer of the required turns for all wires.
This is incorrect, or at least, based on piecemeal experience. Let me fill in some foundation theory.
- CMC cores are the best. The same core materials are used for pulse transformers, which is what GDTs are.
- The permeability is the highest possible (typically 8k-15k for ferrite, nanocrystalline up to 40k). For a CMC, this ensures maximum filtering impedance in the least size and (presumably) material cost. For a GDT, this ensures minimal impact from magnetizing current (which causes droop in the square wave tops, and increases load on the driver).
- Typical inductor* cores (usually #52 or #26 powdered iron, gapped ferrite shapes, or ferrite rods and other open shapes) have effective permeability in the 10-100 range. This loads the driver, and can cause malfunction or failure if the driver can't handle that much magnetizing current (typically, drivers are rated for so-and-so ordinary load current, but a modest fraction of that current in reverse, i.e. through the output transistor body diodes; exceeding that limit causes CMOS latchup and probably destruction).
*I'm using the classification that an inductor is intended to store energy, while a transformer is not. Transformers must have high permeability cores, inductors must have moderate. Note that, by this definition, a flyback transformer is a coupled inductor.
You can calculate how many turns, on what size core, is needed. For square waves, this is:
N = Vpk / (4 * Bmax * Ae * F)
Vpk is the "top" voltage, Bmax is the maximum flux density (determined by saturation or heating, usually under 0.3T), Ae is the effective core cross-sectional area, and F is switching frequency.
You need more turns around a thinner core, which tends to give longer wire length (= more resistance and inductance). You are quite correct that bigger is usually better (to a point, of course). :)
A wider core is also better, though these are hard to come by in toroid format. The best alternative: swap wire and core, so the core wraps around a ring shaped coil instead. These are called pot ('P') cores. They have excellent A_L (inductance / turn^2) and Ae, even for modest materials (mu_r ~ 5000 is oddly common to see). They're also much easier to wind, just wrap the wire on the bobbin and close the core around it. Downside, you're probably going to have to buy them (a few bucks each).
Incidentally, the factor of 4 becomes 4.44 when sine wave Vrms is used. To be exact, it's a factor of pi*sqrt(2).
Likewise, you can estimate the leakage inductance of the transformer, by taking the wire length used (assuming a transmission line transformer construction, as is under discussion here), and multiplying it by the inductance factor of the wire (usually ~0.6uH/m for twisted pair).
My prefered option is to use really thin wire, like from a cheap syncronous motor (less than hair thin) and to make several lengths of litz wire from it.
This is a good way to make Litz, yes. Litz probably isn't needed for GDTs, unless you're needing a serious amount of power.
Incidentally, Litz allows magnetic field to penetrate it -- that's why it has lower AC resistance -- but this increases the transmission line impedance (increased leakage inductance) of the transformer! The space between primary and secondary is where leakage field is stored.
GDTs usually need a much lower impedance, than the usual transmission line impedance, so -- whether or not you use Litz -- you should use several twisted pairs, wired in parallel, for each winding. (And if you do use Litz, you may need to use additional pairs!)
The actual impedance required, for hardest drive strength, is equal to the gate resistance of the transistors and driver, plus whatever you've added to the circuit. So, for small transistors, 10 ohms is a good start, but it could be lower. For big industrial modules, it could be 1 or 2 ohms! At this point you start to ask if it's better to use a separate isolated circuit (which indeed is what's usually done in that case :) ).
In case you want to try with simple wire first salvage some line filter cores for their wire or better still use the (copper) wire from the primary of a MOT.
Clamp one end in a vice or twist it around something sturdy and with a tight grip run an old rag along the wire - this will remove any leftovers from the varnish so it won't act like needles when you wind it.
Use a pair of pliers or wrap the wire around a grip to slightly pull on it until perfectly straight.
Cut to the estimated lenght you need and wind it with care to prevent kinks.
Again wind all strands at once and in parallel to have the inner surface of the core covered while leaving small gaps on the outside.
Be careful with salvaged wire, especially where isolation is needed: it can be nicked, cracked and flaked, and this can be hard to see, especially with clear/gold enamel. Keep that in mind, and insulate generously if you can (yellow polyester tape is great stuff!). You'll be alright. :)
While CAT5 and such cable isn't normally rated for much voltage, I don't mind using it for quite a bit more. I definitely wouldn't consider it reinforced insulation (which is to literally say: I wouldn't trust it with my life!), but functional or basic even, can't be too bad for one-offs.
Another way out is to use mosfets to provide the gate signals.
For fast switching og IGBT's you need high currents and a fast enough driver, mosfets can do this quite good.
If your inverter layout is tight, and peak load current is modest, consider bootstrap gate drivers (which, of course, have MOSFETs internally). :)
They don't scale up well, due to limitations on peak current capacity and transient voltage. Particularly undershoot, when the switch node transitions from high to low and bounces below logic GND, where the internal level translation circuit momentarily stops working. Hence, tight layout and modest load current (maybe 10A tops, for a layout like pictured above?).
Tim
-
@Hydron
I also thought that the 590ns reading is the rise time,but I did not want to say anything maybe incorrect :P.I am using MUR1560 in parallel with 440V TVS diodes as protection even just with SSTC operation.Here is the driver stage schematic:
(https://i.imgur.com/3xlZqfZ.png)
@Downunder35m
I do not have any other power supply at 24V.I'm planning to build a single MOSFET SMPSU in the future accordingly to this schematic: http://danyk.cz/impulz3.png or rewind my big 50V transformer for lower voltage.
Anyways I put the whole driver back together.The final waveforms:
(https://i.imgur.com/ZZU9BeD.gif)
The slow rise and fall time causes the IGBTs in the halfbridges to turn on at the same time for a brief moment,as you can see here:
(https://i.imgur.com/lWKZkE4.jpg)
The bridge output at 20V input,with no load:
(https://i.imgur.com/EZQamRw.gif)
What is your opinion about it?Is there any way to test the inverter if it works OK without connecting it to the primary,or is it working already good enough for a proper test?
-
I think there is some confusion here - the rise-time is NOT 40ns
I will go dig a hole and throw myself in it right away, I stared myself so blind on that label :-[
-
You could easily go lower than 4.8ohm gate resistances without a problem
In your case though I suggest you to switch to a dead time adding circuit(that is only if you really mind the overshoot)
The though this overshoot might just be an artifact of low voltage testing
Use the full voltage on the bridge and let us know the results
-
@ Offline T3sl4co1l : Thanks for the detailed info!
I would be really nice if we could collect such replies or informations somewhere.
My so called knowledge with RF, inductors and Tesla stuff in general all started when I was still in school.
Now, 40 years later a lot of what I learned at the time seems to be far less relevant, at least for our purposes here.
When I learned how to make my first radio transmitters the old blokes always insisted to use Litz wire whenever possible and practical, same for twisting wires that carry more than a supply voltage.
I suggested to use MOT wire for salvage because I made really good experiences with it.
As a rule of thumb I first check if I can solder through the insulation at the max of my soldering station.
If that works than I use the wire for normal connections on ciruit boards, high power but only mains voltage transformer applications and so on.
Only if it is a real pain in the behind to solder it directly I use it for HV.
And just becuse I can I often double check after the entire lenght is stored on a reel or similar.
For this I use a bth of water with some baking soda dissolved.
Submere the wire but leave the ends out.
Put finger in the water and use a spark igniter from an old lighter to spark ont the free wire end.
No shock and all good, shock and someone was not careful enough when getting it off the old core....
@ Laci: Someone will certianly shoot me virtually for it but anyway:
When I was at my peak with simple induction heating I literally had a small bucket full of fried mosfets.
Most were salvaged so no big loss at the time but when I went to more powerful versions I decided to avoid the digital smell of failure.
Due to the design of my circuit there was a high chance of latching if a diode or resistor was only slightly out of specs.
And a lot of amps going through both mosfets usually caused the instant death syndrome.
I simply replace my entire LC tank with a beefy 10W resistor and for the supply my lab power supply with a very low current set on the limiter.
Just enough for the resistor.
If the circuit was bad the mosfets would survive most of the time because of the very limited current available from the lab power supply.
But I don't see much that would be reason of concern in your case, so why not start testing under real conditions with a low input voltage?
-
I would try using a lossy ferrite bead for the gate resistor or in addition to a small gate resistance. The goal is to have the bead be lossy for the high frequency AC overshoot but not affect the relatively slow turn on/off gate signal. I have had good results with them. Here is an app note: https://www.fairchildsemi.com/application-notes/AN/AN-9005.pdf (https://www.fairchildsemi.com/application-notes/AN/AN-9005.pdf)
-
I would try using a lossy ferrite bead for the gate resistor or in addition to a small gate resistance.
That is actually a really great idea!
-
Ferrite beads are nice: they saturate at high peak currents, so the gate rise/fall time isn't compromised much if at all. The signal is delayed slightly (a few ns with typical drivers), overshoot is usually modest (the inductance doesn't store much energy relative to the much larger gate capacitance -- when it is larger, anyway, so mind this will be different for small transistors!), and the steady-state dampening is excellent (preventing oscillation).
The downsides are:
1. Because of saturation, you may still get singing during hard switching conditions.
2. Because of operating at high Bmax, core loss is quite high. FBs get impractical above maybe 300kHz (for ~20A 500V transistors) or lower with larger transistors. What happens is, temperature rises above the Curie point, impedance tanks, and the FB stops FBing. It thermally self-limits, exactly as a Metcal soldering iron tip does (if at lower temperatures). This is most significant for NiZn ferrites, which have a lower Tc than MnZn, but in that case, mind the reduced current rating and higher operating temperature!
Tim
-
The downsides are:
1. Because of saturation, you may still get singing during hard switching conditions.
2. Because of operating at high Bmax, core loss is quite high. FBs get impractical above maybe 300kHz (for ~20A 500V transistors) or lower with larger transistors. What happens is, temperature rises above the Curie point, impedance tanks, and the FB stops FBing.
This seems a bit broad of a statement. Can you elaborate how you know a bead will be operating in saturation? If it is mainly a question of power ratings, you can parallel a few?
-
The downsides are:
1. Because of saturation, you may still get singing during hard switching conditions.
2. Because of operating at high Bmax, core loss is quite high. FBs get impractical above maybe 300kHz (for ~20A 500V transistors) or lower with larger transistors. What happens is, temperature rises above the Curie point, impedance tanks, and the FB stops FBing.
This seems a bit broad of a statement. Can you elaborate how you know a bead will be operating in saturation? If it is mainly a question of power ratings, you can parallel a few?
Yes. Take a look at Laird's catalog: they have DC bias on almost all their parts. (Other manufacturers have spotty data, and most have no data at all on DC bias.)
https://assets.lairdtech.com/home/brandworld/files/Catalog_EMI%20FILTERING%20RF%200717.pdf
A typical 1206 (390Ω) saturates around 300mA, and is rated for 2A DC. The RLC characteristics are approximately inductive below 10MHz, lossy up to the peak (about equal R and X components (Q ~= 1), which means |Z| increases about as sqrt(F) in that range, or |L| goes as 1/sqrt(F)), and capacitive past the peak. The dominant L part goes away as DC bias goes up, being -30% at a fraction of the DC rating. Hence the curves get higher and peakier as bias goes up. At rated DC bias, it's little more than a scrap of wire, 10s of nH.
This all varies with value, of course, with higher Z and smaller size parts dropping more rapidly with current (e.g., 0603s down to 20mA). Large ones don't really handle that much Isat, even in small values: if you need impedance under DC bias, use a CMC or a real inductor!
Tim
-
Just to clarify for this example we are talking about the MI1206L391R. Is the 300mA saturation a derived value? I did not see it listed but I could have missed it.
When I look at its impedance graph, I see the impedance tends towards 0 below 1MHz. To me this is saying there is little power dissipation below 1Mhz so temperature should not be an issue? The ringing we want to get rid of is in the low to 10s of MHz but the amount of energy in the ringing should be far less than the signal.
I agree the current limit of 2 amps is the limit and this cannot be exceeded. The package is only so big...
I sent an email off to a contact at Laird to ask them for their opinion. I will post the response if I get one.
-
So I finished the building of my new coil,the resonant frequency is around 170kHz as the JavaTC calculated.Today I put everything together and started testing.Sadly it performs as badly as before-nowhere near in comparision with a halfbridge design.The input voltage was more than 50V,but it was drawing over 4A even at 20V in CW mode,with some mm weak arcs.At the 50V supply I had a 4700uF smoothing capacitor,which discharged really fast in interrupted mode,when the supply(transformer)was turned off.I suppose there are huge losses,probably at the points that I marked before.Tomorrow I will make some more test,maybe measuring the inverter output,gate drive waveforms vs input current(through a shunt resistor).
The ferrite bead you are talking about is something like this?
(https://i.imgur.com/0B12PIM.jpg)
I have a similar thing salvaged from an angle grinder,sadly nothing else.
A picture of the current setup:
(https://i.imgur.com/87Al4ZI.jpg)
-
Just to clarify for this example we are talking about the MI1206L391R. Is the 300mA saturation a derived value? I did not see it listed but I could have missed it.
Yes. See where the impedance at 100MHz drops to 70% of the zero bias figure. You'll have to roughly interpolate the curves they give. The real value is somewhere around there.
When I look at its impedance graph, I see the impedance tends towards 0 below 1MHz. To me this is saying there is little power dissipation below 1Mhz so temperature should not be an issue? The ringing we want to get rid of is in the low to 10s of MHz but the amount of energy in the ringing should be far less than the signal.
Ah, but we are driving gates here, so the current waveform is very peaky, the derivative of a square wave (approximately)!
I have a project that runs FDP33N25s (Qg = 37nC typ.) at up to 2MHz, with a small ferrite bead (might've been Fair-Rite #2673000101?) on the gate, and they get warm! Not up to Tc, though. I've cooked (up to Tc) the same size beads, in #43 material, on the S or D leads of STP19NM50s without any load at all (just commutation current: also very peaky, but much sharper, more intense, than gate drive!)
More specifically: an H bridge, hard switching, no load current, 300kHz, 320VDC supply. So, just charging Coss through stray inductance and ferrite beads.
I agree the current limit of 2 amps is the limit and this cannot be exceeded. The package is only so big...
Which, again, is only a thermal rating, for some typical test condition. You're welcome to run hotter, at risk of cracking or oxidation. Or maybe you can't handle the losses, and you need to run less, anyway.
And core loss directly impacts the current rating, just as it does for regular inductors. It might be rated 2A at DC, but 1A when, say, 300mA (RMS) at 1MHz is also applied, or whatever. Or it might be 3A (DC) with generous copper pours, or less with minimal footprints.
I sent an email off to a contact at Laird to ask them for their opinion. I will post the response if I get one.
Cool, will like to hear. :)
Tim
-
hi Laci, I don't know if you got this problem solved or not but what I am seeing in these pictures is shoot through. if it were me I would add dead time that is a little grater than your 560ns rise and fall times like 1us. If this happens with your main bridge it will be undesirable. If you drive each mosfet directly with a driver with 1-3 ohms. I found this
http://omapalvelin.homedns.org/tesla/SSTC/half-bridge.htm
http://www.modularcircuits.com/blog/articles/h-bridge-secrets/h-bridge-control/
http://www.loneoceans.com/labs/ud27/
hope this helps
-
Hi Teravolt!Thanks for the useful informations!The coil is working well yet,I have not tested it at high power,but it is already too noisy to run indoors.I am planning to modify the current oneTesla interrupter to add a mains synced staccato controller to it for long and more quiet sparks.
The waveform is the same as before,a bit high rise and fall time.One thing I figured out is that there is no problem with the gate resistors,since the fall time is also high through the gate schottky diodes.I suppose my ferrite core is not the best in my situation.The deadtime is a great idea,not sure if I have enough empty space on the driver board for some more components.Hopefully the 440V TVS and MUR1560 duo will prevent the IGBTs from dangerous overshoots! :)
-
I don't think there's a need to add the MURs as the IGBTs already come with built-in diodes which are much more optimized for the task at hand
I would not suggest you to go for that
-
I have IRG4PC60U's which do not have built in diode.The 440V TVS diodes should be enough,I just keep them for safety against my high inductance bridge design.:P
-
Oh, oops
Should've read the intro post!
Anyways then I definitely recommend the MURs