QCW questions

[Rafft]

BUT why does connecting a scope probe 'fix' the ramp output? my coil is battery-powered and no connection to ac line (only scope is connected to ac line, for power)

My guess would be a change in grounding rather than a change in ramp shape.  Scope is adding a ground connection path.  If it is ramp related, I'd suspect that without scope ground something in your buck control circuit is getting confused by noise.

come to think of it, heatsink for buck is floating. will add bypass cap to buck GND

David

would the lo-side benefit from a mosfet(as switch) + another diode in parallel? my thinking is the diode will be the 1st one to catch , followed by the fet(so it does not have to carry the bulk of the discharge). would this work?

[Rafft]

update:

 *seems*  Ive forgotten to reconnect the output LC GND to  buck control GND. this connection was on the LO out going to IGBT!  sheeesh! that was the reason why async did NOT work before!

anyways, problem solved! buck now more robust. only high side switch to worry about 

now on to higher coupling! (k=0.59) I hope I dont get flashovers . lol

[davekni]

anyways, problem solved! buck now more robust. only high side switch to worry about 

Great to hear that you've found and fixed the issue!

would the lo-side benefit from a mosfet(as switch) + another diode in parallel? my thinking is the diode will be the 1st one to catch , followed by the fet(so it does not have to carry the bulk of the discharge). would this work?

Presuming I'm understanding correctly, this is often called "synchronous rectification".  An external diode isn't required, but can be useful if lower Vf and faster than the MOSFET's internal body diode.  Synchronous rectification is common in low-voltage converters where diode Vf is a significant fraction of output voltage.  For your buck, I wouldn't bother with synchronous rectification.  Diode(s) alone as you have now should work well, presuming fast diodes and sufficient current capability.

[Rafft]

Hello again

currently im using (slow) IGBTs IRG4PC50FD in full-bridge for the inverter and a slightly modded UD2.1b.

1:32:31/10R for OCD and I have set LM311 comparator to trip at 1.1v (110Amp). ramp from buck ranges from around 4mS to 8mS. anything more than that, its just adding *hiss* and no more spark length. my boost can generate up to 300Vdc. storage cap 6600uF total.

testing some lower taps on my primary coil CAN trigger the ocd. my coil setup loves to run on 450KHz. resonant freq of secondary/topload is 377KHz(15.34mH/8.135pF k=0.374), so basically its running on upper pole.

am I conservative at 110Amps? what could possibly be my 'max' amp setting from these IGBTs?

thanks

[davekni]

am I conservative at 110Amps? what could possibly be my 'max' amp setting from these IGBTs?

I'd guess you could go slightly higher.  Depends on your risk tolerance for fried parts, and on how well phase-lead is working (how significant switching loss is).  Specified peak current capability for IRG4PF50WDPbF is 204A.  I don't see a time duration for that peak.  1ms is typical for such specs.

Consider the case of 200A peak.  Average current for 200A sine wave is 2/PI * 200A = 127.3A.  Each IGBT conducts half the time, so 64A average.  Extracting from graphs, guess of 5V average forward drop for 320W conduction loss.

Switching loss is the harder part to estimate.  Even with "ideal" phase lead, switch-off needs to occur when some current is still flowing to cause H-bridge output voltage transition.  Switch-off also needs to be early enough so that opposite IGBTs turn on before significant current reversal after dead-time.  Highest turn-off loss spec. is at 28A and 25C, 1.38mJ.  No data on turn-off energy vs. temperature.  Also would be hard to get phase-lead perfect enough to be down from 200A to 28A, especially at 450kHz.  I'll take a wild guess at 5mJ for turn-off energy with hot IGBT die and whatever current switch-off occurs at.  That's 450kHz * 5mJ = 2250W per IGBT, much more than conduction loss.  Total for this guess is 2570W.

If IGBT die starts at 50C and rises to 150C, 100C delta, tolerable thermal resistance is 100C / 2570W = 0.039C/W.  Spec. thermal impedance plot hits that at ~250us pulse width.  Since you need at least a couple milliseconds close to limit current, 200A is likely too high.  Of course, this is a wild guess at switch-off loss, which is dominant for your high frequency use of medium-speed IGBTs.  200A could work for short bursts of a standard DRSSTC per this guess calculation.

[Rafft]

David

I'd guess you could go slightly higher.   

thanks! I really like this answer  I was thinking maybe I could push this to 120A-130A . Ive read some old posts that this specific IGBT could go 4x its Ic rating(or was it the Icm x4) but that was for DR and 50KHz.

with regards to your explanation,  this still goes down to using a better (faster) IGBT then. was thinking of using THIS
https://www.google.com/url?sa=t&source=web&rct=j&url=https://uelectronics.com/wp-content/uploads/2021/02/MBQ60T65PES-Datasheet.pdf&ved=2ahUKEwiAouKrosf2AhWRslYBHZb9CnsQFnoECDEQAQ&usg=AOvVaw271Wb2Nv5obmTWlhJivFyC

hope this will be better

[davekni]

thanks! I really like this answer  I was thinking maybe I could push this to 120A-130A . Ive read some old posts that this specific IGBT could go 4x its Ic rating(or was it the Icm x4) but that was for DR and 50KHz.

For low frequency, my rule-of-thumb is that IGBTs fry at 4x continuous current rating or 2x peak current rating.  Margin is needed below frying current.  I've ran destructive testing on two of my STGW60H65DRF parts, getting a ways past that rule-of-thumb.  Ran second one to 650A peak, more than 2x it's 240A peak rating.  Diode fried first at 600A peak, but still over the diode's 240A peak rating.  This was at 80kHz to match my DRSSTC.  Short ring-down bursts, so likely to fry at a bit lower current in real DRSSTC use.

with regards to your explanation,  this still goes down to using a better (faster) IGBT then. was thinking of using THIS
https://www.google.com/url?sa=t&source=web&rct=j&url=https://uelectronics.com/wp-content/uploads/2021/02/MBQ60T65PES-Datasheet.pdf&ved=2ahUKEwiAouKrosf2AhWRslYBHZb9CnsQFnoECDEQAQ&usg=AOvVaw271Wb2Nv5obmTWlhJivFyC

hope this will be better

Yes, spec's look great!  Very low Eoff and very low gate charge.  (And Eoff is specified at 60A rather than 28A as your existing part.)  Have you found a place to purchase these?  A quick look at oemstrade.com shows little stock availability.  I may buy some if/when they are available.

[Rafft]

Have you found a place to purchase these?

yes. its sold by an online company/seller who does COD.

ordered a few. I do hope I get the real deal.

edit 1:
and from your last post, IRG4PF50WD(900v device) is different from IRG4PC50FD(600v Ic=70A@25degC 39A@100degC Icm=280A and much slower than PF50WD)

[Rafft]

ok. making progress! 

tried 'high impedance' PRImary setup. and after a few more calculations regarding HOW to tune it(almost), im Loving the result!  old logs/build notes really help!

didnt think(!) I would get a goooood spark output with LESSER amp draw. coil freq=450KHz

this is from 3rd CT 1:100T/1R burden. 100A per division. blue trace(Iprim) yellow is full-bridge output. its about 50Amps only. almost half of my usual.


my limiter is currently set to 1.1V(110A) so I guess im -safe- to push it farther a bit to maybe 150A 

bridge output does not look 'sharp' at the top edge. nothing to worry, Ive just overdriven the ramp gen to do THAT 

[Rafft]

hi guys

how is it easy(or hard) to implement a frequency change-over switch for the UD2.x?  an adjustable free-running oscillator will initiate the spark then natural feedback would take over in a few clock cycles. doable with logic gates? arduino?

[davekni]

Haven't tried this for QCW, but one option is to modify UD2.7 to be self-oscillating.  Here's a schematic of my preferred variation:
https://highvoltageforum.net/index.php?topic=1336.msg9894#msg9894

A variation can be done with fewer modifications to UD2.7 as described in third paragraph of this post:
https://highvoltageforum.net/index.php?topic=1373.msg10197#msg10197

These are intended for more reliable DRSSTC startup in cases where feedback is weak or initial H-Bridge state is wrong (same output voltage already as what the driver initially forces after enable starts).  However, by adjusting component values, the self-oscillation frequency could be made to match a QCW upper-pole frequency and made a bit stronger so that feedback doesn't overtake it for several cycles.  If you are interested in trying this approach, I'd be happy to help further with figuring out best component values etc.  Would be an experiment, not a guaranteed solution.

[Rafft]

Hi David

remove R7(100k), add 50k and adjust C33. seems doable. what does SV1 mean? short the inductor? if so, no more phase lead?

If I decide to do this, should I 'tune' my coil near lower pole(as calculated together with coupling coeff)?

so what happens is coil(tuned to lower pole) will start oscillating at upper pole, then afrer a few cycles, oscillate to lower pole, right?

[davekni]

Part number references are for this UD2.7 schematic:
https://www.loneoceans.com/labs/ud27/UD27Cschematic.png

SV1 pins 2-3 are shorted to include phase-lead, pins 1-2 for no phase lead.  If the jumper is left off, there is no burden resistor for feedback CT, so it won't work correctly.  Sorry for the confusion.  I should have listed the schematic reference.

If I decide to do this, should I 'tune' my coil near lower pole(as calculated together with coupling coeff)?

Not sure exactly what this means.  I think that QCW designs using UD2.7 without any special features need to tune uncoupled primary frequency slightly higher than uncoupled secondary frequency.  That way UD2.7 tends to start at upper pole (presuming sufficient phase lead).  With a PLL design or UD3.x, it is possible to force upper pole operation even with primary frequency tuned lower than secondary.  I believe this improves performance.  Arc loading lowers secondary resonance, keeping frequencies close over entire arc growth.  That should improve performance.

(One qualifier on any QCW suggestions I offer:  My only personal QCW experience is my low-frequency QCW experiment.  It was more like a QCW SSTC with coupling of 0.9, way higher than other designs.  All other QCW knowledge is from simulations and analyzing designs posted here.)

so what happens is coil(tuned to lower pole) will start oscillating at upper pole, then afrer a few cycles, oscillate to lower pole, right?

I'm guessing that "tuned to lower pole" means primary tuned lower than secondary, in which case UD2.7 starts at lower pole.  I think the goal is to keep operation at upper pole for the entire ramp.  (Both pole frequencies will drop with arc loading.)  I am not at all certain that UD2.7 will stay at upper pole once feedback takes over.  Hopefully someone else can offer insight here.  (Or perhaps I can find time in the next week or two to run simulations.)

If we make the UD2.7 self-oscillation strong enough, perhaps it can still influence frequency even after feedback is strong enough to provide most of the signal.  Don't know.  Should have a better chance of this if D1 and D2 are silicon signal diodes (1N4148 or similar) rather than schottky diodes, and R2 increased to 5-7k.  With that, I'm guessing (without simulation yet) that C33 value of ~560-680pF would result in 450kHz operation before feedback.  (Test with no power to H-bridge.  Measure frequency on UD2.7 output or GDT or IGBT gate.)

Good luck!  Remember, no pressure to try this.  Only if you feel like experimenting.  I am quite curious how it goes if you do try.  Or wait a bit for me to try simulations first.

[Rafft]

David
primary current

I'm guessing that "tuned to lower pole" means primary tuned lower than secondary, in which case UD2.7 starts at lower pole.  I think the goal is to keep operation at upper pole for the entire ramp.  (Both pole frequencies will drop with arc loading.)  I am not at all certain that UD2.7 will stay at upper pole once feedback takes over.  Hopefully someone else can offer insight here.  (Or perhaps I can find time in the next week or two to run simulations.)

yes. exactly. kindly check my javaTC file. MY coil resonates from 450KHz(ramp start) going down to 431KHz(ramp TOP). 5mS total, but not including ramp-down(roughly 1mS). I tried higher tap (13.2uH) and it went down to lower pole (316KHz-312KHz @ 45Apk only, low spark output). btw, Im using the MBQ60T65PES (full bridge - 4pcs IGBT). I doubt they are the real deal(they look ugly). not sure also if this is(was) developed in china. havent pushed it yet, just limiting to 120Apk.

If we make the UD2.7 self-oscillation strong enough, perhaps it can still influence frequency even after feedback is strong enough to provide most of the signal.  Don't know.  Should have a better chance of this if D1 and D2 are silicon signal diodes (1N4148 or similar) rather than schottky diodes, and R2 increased to 5-7k.  With that, I'm guessing (without simulation yet) that C33 value of ~560-680pF would result in 450kHz operation before feedback.  (Test with no power to H-bridge.  Measure frequency on UD2.7 output or GDT or IGBT gate.)

I will see what I can do. my UD is built on pad-per-hole board. almost every pcb trace, is a component leg.  im using MAX913 for the comparator.

what Im seeing here is that start-up is 450Khz(this is close to 430KHz=big sparks), coil (feedback to be 'lower pole', as set by PRI LC), would the output go from big spark to 'smaller' since it will get feedback from lower pole freq?

ok let me wait for your SIMs 

my coil currently (javaTC)

Code: [Select]

units=0,
ambient=1,
s_ws=0,
s_Al=0,
p_ws=1,
p_Al=0,
p_ribbon=0,
temp=20,
g_radius=0,
w_radius=0,
ceil_height=0,
s_radius1=1.6,
s_radius2=1.6,
s_height1=9.25,
s_height2=13.7,
s_turn=626.8,
s_wd=33,
p_radius1=2.9,
p_radius2=2.8,
p_height1=9.25,
p_height2=10.33,
p_turn=6.834,
p_wd=0.079,
p_vwidth=0,
p_rthick=0,
Cp_uF=0.0142,
Lead_Length=0,
Lead_Diameter=0,
desired_k=0,
t.inner=1.3,
t.outer=6.4,
t.height=14.7,
TT=true,
TG=false,
t.inner=1.3,
t.outer=6.8,
t.height=16,
TT=true,
TG=false,
x_Vin=0,
x_Vout=0,
x_Iout=0,
x_Hz=0,
x_Vadjust=0,
x_ballast=0,
rsg_ELS=0,
rsg_ELR=0,
rsg_rpm=0,
rsg_disc_D=0,
rsg_ELR_D=0,
rsg_ELS_D=0,
stat_EL=0,
stat_EL_D=0,
stat_gap=0,
SPE=true,
RGE=false

screenshot grab of the igbt internals (taken from buyer comments, same seller I got these igbt from)... small? bad? legit?




-Ralph

[davekni]

Ralph,

yes. exactly. kindly check my javaTC file. MY coil resonates from 450KHz(ramp start) going down to 431KHz(ramp TOP).

Simulating with your JavaTC results starts at 470kHz.  That discrepancy is within reason.  Perhaps a bit more top-load capacitance due to non-infinite distance to surroundings or some such reason.

I will see what I can do. my UD is built on pad-per-hole board. almost every pcb trace, is a component leg. my C33 is 1nF(is this ok?), still based on orig schema.

ok let me wait for your SIMs 

Short answer:  Probably not reasonable to try with your driver.  I did manage to get simulation cases to work, but with significant circuit changes.  Would likely require experimentation (adjusting values) to get it functioning in practice.  In case it is of interest, here is a working simulation example:

 [ You are not allowed to view attachments ]
Attachment ended up at end of this post.

Required dramatically reducing feedback CT signal strength and increasing comparitor positive-feedback amplitude, along with adjusting self-oscillation frequency to slightly above upper pole frequency.  With those changes, remained at upper pole with change of primary MMC from 14.2nF to 17nF, so primary frequency below unloaded secondary frequency.  Had some cases working even at 20nF MMC, but tricky there.  Narrow range of component values that worked.

BTW, B1 is a voltage-controlled voltage source.  It is replacing (for simulation) gate driver chip through IGBTs.  Since it has zero delay, phase lead is set very low here.  Would need to be increased to match real circuit delays.

and im using MAX913 for the comparator.

Should be fine for normal UD2.x, but would make above changes even more tricky.  Has higher input bias current and lower output voltage swing than TL3116 , affecting values needed.

I doubt they are the real deal(they look ugly).

screenshot grab of the igbt internals (taken from buyer comments, same seller I got these igbt from)... small? bad? legit?

Certainly can't tell for sure, but could be genuine.  Would be hard to fit much larger die in a TO247 package.  (The upper die closer to the gate lead should be the IGBT and the lower one the anti-parallel diode.)
MagnaChip must sell directly to equipment manufacturers mostly.  I find relatively little about the company and few of their parts in normal electronics distribution.  Appears to be primarily a South-Korean company, though not officially incorporated there.  Quite possible that IBGT/diode die are made there and packaged in China.  Could even be die made in China by or for MagnaChip.  I ordered some from EBay (from China).  AliExpress doesn't work for me for some reason.  Cheaper there, though.
In online pictures I noticed that some parts have numbers or letters molded into the two small disks on package front, while others are blank.  That could be some indication of genuine or not.  I don't know.
Next similar but smaller MagnaChip die is MBQ50T65DSC.  Slower IGBT with higher gate capacitance.  Do you have a meter than can read capacitance?  Try measuring gate capacitance to drain/source shorted together.
Perhaps the most likely scenario is that these IGBTs are rejects from the packaging line (or parts from molding machine setup runs).  In that case they may have genuine die but not complete marking or clean packaging.  Could be more moisture sensitive.  Could have higher failure rate if die-attach is reason for reject (causing hot-spots within die).  Since MagnaChip parts are not generally available from normal sources, they seem like unlikely parts to counterfeit.  Anyone bothering to counterfeit parts would presumably be interested in faking popular parts from well-known companies.

Have fun with your coil!

[Rafft]

David

initial reply(using only phone)
using my store-bought LC meter.
MBQ60T65PES G-C 6nF G-E 4.17nF G-C&E(shorted) 8nF

using my diy LC meter(proven to be more accurate at small values)
MBQ60T65PES G-C 3.479nF G-E 4.193nF G-C&E(shorted) 6.25nF

just for comparison: diy LC
IRG4PF50WD(1200v device) 4.89nF/4.847nF/6.983nF
IRG4PC50FD(slower 600v device) 5.05nF/5.629nF/7.35nF

so, is this MBQ fake?

my actual IGBTs


start ramp 454KHz


ramp top 431KHz. also I think I have a problem with that GDT core(material). output doesnt look good enough(?)


yellow= bridge output blue=Iprim(100A per div)


I have been also playing with the igbt gate resistor values. Im using just 1 core for gdt for the 4 fullbridge IGBT. above scopeshots was from recent 2R2(and 5819).

here is from 4R7


and 15R (what are those extra spikes again? body diode?)


thanks for the sim. 

I do have another topload. not a uniform 3.4"×9.6" because it was formed from a 4" diameter aluminum ducting. one thing I noticed using THAT torroid was, it was not that straight of a spark(compared to the smaller 2 stacked) even with oscillation above 400KHz.

Ralph

[Rafft]

David

UPDATE #1

as per your posted schematic. as I have mentioned before, MAX913. its all I have for a fast comparator. using 10R and 24uH for the filter. still have to play around with 22R 33R 47R in actual.




with ALL the parts connected, 446KHz, else 2.4MHz with just the comp and some passives. output taken from pin7. this should be at 470KHz, right? I will try smaller value for C33, 200pF - 330pF maybe. btw for clock circuits, MLCC not good? maybe get a regular ceramic cap in there?


-Ralph

[davekni]

Ralph,

so, is this MBQ fake?

Based on capacitance measurements, I'd guess fake, but can't be certain.  That's presuming the datasheet for MBQ60T65PES that you posted is accurate.  Figure 7 shows Cies (G to CE shorted) as being about 3.5nF at 0V.  Your measurement of 6.25nF is 78% high.  Other two parts you measured are about 18% and 22% above what their data sheet graphs show.  These data sheet graphs are typical values.  Thus this is not proof of counterfeit.  It is quite possible that MagnaChip couldn't meet their initial values and updated a datasheet.  Hard to tell since little information shows up on the web.  Their datasheet capacitance values and gate charge are atypically low, one of the reasons I was impressed with the part.  Don't see any obvious evidence of counterfeit in the images you posted.

ramp top 431KHz. also I think I have a problem with that GDT core(material). output doesnt look good enough(?)

My guess is scope ground lead inductance or wire inductance issues - emitter current across wiring inductance being superimposed on Vge.  I run into the same issue when scoping Vge at high currents.  Switching looks cleanest in this scope capture.

and 15R (what are those extra spikes again? body diode?)

This is exactly what happens when phase lead is not quite sufficient.  Higher gate resistance slows turn-on, so would require more phase lead to allow turn-on to occur by zero-current time.  Better to stay with lower gate R rather than increased phase lead, especially at your high frequency.  Good evidence that phase lead is about right for lower gate R.  The 4R7 traces show just the beginning hint of triple-transitions.  That is a fine level to have.  (I've posted here somewhere a couple times explanations about why marginal phase lead causes triple-transitions on bridge output.  Even less phase lead does not produce triple transitions, just hard switching spikes.)

with ALL the parts connected, 446KHz, else 2.4MHz with just the comp and some passives. output taken from pin7. this should be at 470KHz, right? I will try smaller value for C33, 200pF - 330pF maybe. btw for clock circuits, MLCC not good? maybe get a regular ceramic cap in there?

Is 446kHz with H-bridge powered, so entire feedback loop?  2.4MHz seems quite high.  What parts were connected for that.  BTW, the high duty cycle (longer positive) is likely due to comparitor's lower output voltage and higher input bias current.  Can be fixed with value changes.  If you want to experiment, it requires high local feedback amplitude (at comparitor) and low primary current feedback amplitude.  That is the change to silicon diodes and 5-7k (instead of 1k) on input, 10ohm burden resistor, and 500:1 CT ratio.

No inherent issue with MLCCs.  For timing, look for NP0 or C0G dielectric.  Those designations indicate nominally no temperature or voltage coefficient to capacitance (stable capacitance value).  Usually fairly high Q (low ESR) too.  Z5U and Y5V dielectric caps are worthless (exaggerating slightly).  X5R and X7R dielectrics have gotten worse over the years as part sizes shrink.  Some lose 90% of capacitance at rated DC voltage.  Others aren't quite that bad.  For precision, stay with NP0/C0G (or film capacitors).

Do you have a schematic for the little board you show in the last image?  Then we can discus specific part values by reference designator. 

[Rafft]

David

  I'd guess fake

kinda expecting that especially that they seem 'çheap', about 10USD for 5pcs. the other 2 IGBTs 4PF50WD(this is genuine) & 4PC50FD(bought locally YEARS ago, perhaps genuine as well(?)).

My guess is scope ground lead inductance or wire inductance issues - emitter current across wiring inductance being superimposed on Vge.  I run into the same issue when scoping Vge at high currents.  Switching looks cleanest in this scope capture.

This is exactly what happens when phase lead is not quite sufficient.  Higher gate resistance slows turn-on, so would require more phase lead to allow turn-on to occur by zero-current time.  Better to stay with lower gate R rather than increased phase lead, especially at your high frequency.  Good evidence that phase lead is about right for lower gate R.  The 4R7 traces show just the beginning hint of triple-transitions.  That is a fine level to have.  (I've posted here somewhere a couple times explanations about why marginal phase lead causes triple-transitions on bridge output.  Even less phase lead does not produce triple transitions, just hard switching spikes.)

Ooops forgot to mention, that trace was at bridge output, not on igbt gate/s. sorry.
I was also expecting a cleaner output since I have used a supposedly fast IGBT  so yeah, I will just keep gate resistors at 2R2.

Is 446kHz with H-bridge powered, so entire feedback loop?  2.4MHz seems quite high.  What parts were connected for that.  BTW, the high duty cycle (longer positive) is likely due to comparitor's lower output voltage and higher input bias current.  Can be fixed with value changes.  If you want to experiment, it requires high local feedback amplitude (at comparitor) and low primary current feedback amplitude.  That is the change to silicon diodes and 5-7k (instead of 1k) on input, 10ohm burden resistor, and 500:1 CT ratio.

446KHz is just the comparator board, not with whole system.  it oscillated at 2.4MHz with just MAX913/50K feedback/1k+1k mid on pin#2/470R+470p. once I connected the rest of the components(all in your schematic, oscillation lowered down to 446KHz(but still not installed in the coil). my feedback CT is 81:1 (9:9:1). indeed the output on pin#7 has higher on-time than on inverted output pin#8.

Do you have a schematic for the little board you show in the last image?  Then we can discus specific part values by reference designator.

lets use this for reference 



-Ralph

[Rafft]

UPDATE #2

done 50/50% duty ratio and adjustable(to some extent) frequency & set to 470KHz



edit:
with the circuit running(bench test), without CT present, I touched the 'ct terminals' with my bare hands & sure enough, output frequency is influenced(a bit)