Phase lead help

[acobaugh]

I'm in the process of final integration testing of my model 82D coil, seen here: https://labs.cobaugh.io/hv/tesla/82D/

Previously I was using the same bridge and driver in my 81D coil, which can be seen here: https://labs.cobaugh.io/hv/tesla/81D/

The fundamental difference in these two is the resonant frequency. 81D was tuned at around 67kHz, whereas 82D is tuned at around 38kHz. I was using a 7M3-393 variable inductor to set phase lead, and got near perfect ZVS with barely any noticeable overshoot, at least that I can see with my scope. However, at 38kHz, I'm having problems getting any change in the voltage spikes as I adjust the inductor. I started at with the 7M3-393, and I thought I was headed in the right direction by backing out the slug (less inductance), but I ran out of adjustment. I seemed to find a minimum in spikes with the slug all the way in and bottomed out, except for a single spike at the very last half cycle (running about 120uS before hitting OCD). I just tried a 7M3-563 tonight, and adjusting it in/out didn't seem to have much effect. There's a minimum of spikes, except for that last half cycle. Scope shot is shown below.

What I'm asking for is help in deciding which way I might have to go wrt inductance. Assuming same bridge and driver, as one lowers the Fres, how would you expect the required phase lead inductance to change? I'm mostly concerned that there doesn't seem to be a noticeable sweet spot, and adjusting the inductor doesn't seem to make a huge difference. For reference, I'm using a micsig differential probe. I notice the same spike on the last half cycle on both IGBT half bridges, and if I get the pulse width long enough, I can sometimes get it on the last two half cycles.

Thoughts?

By the way, I ran this coil a couple of weeks ago outside (completely out of head room in my shop), and I was hitting ~10ft consistently in burst mode until my remote DC power supply flashed over internally (mistakes were made, correcting those now). Magnetic breakers tripped and everything survived. Now I want to get this thing dialed in perfectly to see if I can break 12 ft. The arcs were only limited by the distance from breakout point to the ground.

[davekni]

It appears that phase is changing with IGBT current.  My suspicion is that the phase-lead inductor core may be saturating.  That would explain why adjusting it has little effect.  Scoping different points along the circuit while zoomed-in at a zero-crossing will show where phase lead and lag are accumulating, from CT feedback through gate waveforms.  Another possibility is IGBT delay depending on current.  (I've read that some old IGBTs had long tail currents after most of the current had turned off.  Doubt that's your issue.)  Your technique of evaluating phase lead by IGBT voltage spikes is effective, often more sensitive than measuring phase directly from scope waveforms.

1200A from a CM300 is impressive.  Seems like that would be on the edge of failure.  My limited experience is that IGBTs at high gate voltages can handle 2x rated peak current (4x rated average current), but not much more.  That's right at 1200A for CM300.

[Mads Barnkob]

Have you made sure that probe skew between channels is ruled out? There is a good guide on how to adjust this in the attached PDF here: https://highvoltageforum.net/index.php?topic=111.0

[acobaugh]

Have you made sure that probe skew between channels is ruled out? There is a good guide on how to adjust this in the attached PDF here: https://highvoltageforum.net/index.php?topic=111.0

Only using a single scope channel, with a micsig dp10013 differential probe. Maybe the rise time of the probe isn't fast enough to see the transients? I was able to see them with this scope (rigol 1054z) before with just regular probes, running the system at low voltage with a transformer-isolated bridge. I can't easily do that with this new setup, though.

And I don't think 1200A is asking too much of these. Many other people have pushed these things even harder.

It's still perplexing to me, this variable inductor situation. The only thing that changed that could affect the phase lead is the resonant frequency. Bridge is otherwise the same as it was a year ago when I first set phase lead on this setup. I only bumped the OCD limit from 800A to 1000A after tweaking phase lead a bit.

[davekni]

That 10013 probe is plenty fast enough.  Barely meets its 100MHz spec, but that's plenty.  If you set the 5MHz mode, then perhaps the transients would be filtered out to some degree.

As long as you have one stable trigger channel, it works well to probe other points one-at-a-time.  If your controller is still isolated (not tied to line voltage), then I'd trigger on one of the GDT primary leads.  Since your scope has 4 channels, I'd use another for current-sense.  Then walk through the circuit with your DP10013.  I'd expand the view around the transition where the glitch shows up, and at an early low-current transition for comparison.  Somewhere there's a difference between the low and high current transitions.

I'm guessing that H-Bridge output spikes occur at times with your higher-frequency coil as well.  I haven't thought of any mechanisms that would make this current-dependent phase change significantly with frequency.

[acobaugh]

That 10013 probe is plenty fast enough.  Barely meets its 100MHz spec, but that's plenty.  If you set the 5MHz mode, then perhaps the transients would be filtered out to some degree.

As long as you have one stable trigger channel, it works well to probe other points one-at-a-time.  If your controller is still isolated (not tied to line voltage), then I'd trigger on one of the GDT primary leads.  Since your scope has 4 channels, I'd use another for current-sense.  Then walk through the circuit with your DP10013.  I'd expand the view around the transition where the glitch shows up, and at an early low-current transition for comparison.  Somewhere there's a difference between the low and high current transitions.

I'm guessing that H-Bridge output spikes occur at times with your higher-frequency coil as well.  I haven't thought of any mechanisms that would make this current-dependent phase change significantly with frequency.

I did some sanity checking on the differential probe. I used a tek p6009 100x probe referenced to ground to observe the exact same transients at the exact same places in the cycle, so it would appear the micsig is able to catch at least these transients.

Thinking about this some more, maybe this bridge is so well-layed-out that transients are mostly mitigated at this lower frequency through layout and , and the transients towards OCD cutoff really phase lead inductor saturation? I have the standard 5.1ohm burden resistor in place, and my fb/ocd CTs are ~1000:1 (50:20:1 cascaded).

I also noticed that C33 is only 1nF on my board (at least from what I can tell by looking at my invoice for that order!). Seems this needs to be changed to 2.2nF, Gao provides examples of what happens if too large of a value is used, but doesn't show what happens when it's too small.

I want to shorten the leads for the ocd/fb CTs (they were further away from the driver in the other design), and try swapping out C33 to see if that makes any difference.

[plasma]

If you plot a phaser diagram with 5.1ohm X for 45 degrees you would want impendence of the inductor at 38khz on the Y axis to be about 5.1ohm.

[Hydron]

The micsig probe has it's issues as Dave said, but shouldn't be missing anything in your use (if anything it makes stuff look worse - there is a nasty lead resonance at ~70MHz).

If you're using a voltage doubler, you can also scope between RF ground (which should be connected to the AC supply by a Y cap, and protective earth and neutral=doubler-mid-rail by the house wiring, giving you the high and low frequency return paths) and one of the bridge outputs if you have a 100x probe.

I wouldn't get too worried about spikes anyway unless they're getting close to the Vce rating of the bricks.

[acobaugh]

The micsig probe has it's issues as Dave said, but shouldn't be missing anything in your use (if anything it makes stuff look worse - there is a nasty lead resonance at ~70MHz).

If you're using a voltage doubler, you can also scope between RF ground (which should be connected to the AC supply by a Y cap, and protective earth and neutral=doubler-mid-rail by the house wiring, giving you the high and low frequency return paths) and one of the bridge outputs if you have a 100x probe.

I wouldn't get too worried about spikes anyway unless they're getting close to the Vce rating of the bricks.

Yep, I compared the micsig to a 100x tek p6009 probe, and the spikes appeared in the same places with the same magnitude.

I did zoom in to some of the crossings. There is definitely a significant, ~1.5uS phase shift in the very last half cycle before the ocd is triggered. The crossings leading up to that were all on the order of 10s of ns, and I could vary those with the phase lead inductor a noticeable amount, so I know it's at least doing something.

The last test I ran today was by cutting the phase lead out of the circuit. The large spike and phase shift on the last 1/2 cycle is still there, even without phase lead. The earlier crossings all had spikes, but they were quite a bit smaller, and honestly not big enough that I would ever worry about them. This bridge just doesn't seem to ring at 38kHz like it did at 70kHz.

That being said, I do want to figure out why there's this consistent phase shift in the last 1 or 2 1/2 cycles. I think I've ruled out phase lead inductor saturation, as it happens without phase lead enabled. Could this be caused by a too small C33?

The way my base and wiring is done makes it difficult to partially pull out the drawer to access things with everything running, and the further out the drawer is, the less air flow I have across the heatsink. So, it's not so simple to go probing the driver with the thing running. So, I'm going to try increasing C33 first to see if that has an effect, then start probing around. I might also try back-probing the feedback CT connector and compare that to the bridge output and primary current, to see if there's anything interesting going on there. That being said, my current probe is made exactly the same way as the feedback and ocd CTs, and those traces always look clean.

[davekni]

"I also noticed that C33 is only 1nF on my board (at least from what I can tell by looking at my invoice for that order!). Seems this needs to be changed to 2.2nF, Gao provides examples of what happens if too large of a value is used, but doesn't show what happens when it's too small."

Presuming the implementation you reference has the same reference designaters as the copy I downloaded, then C33 provides momentary hysteresis after each current zero-crossing.  A larger value is a good idea for lower frequency, protecting for glitches farther into each half-cycle.  It won't affect phase lead significantly until the time-constant is a significant fraction of a half-cycle.  So, I don't think you will notice any difference between 1nF and 2.2nF.  By 4.7nF you might notice increased phase lag.

Yes, good low-inductance bridge layout (snubbers and bulk caps to IGBT bricks) will make switching spikes shorter.  There's always some inductance internal to the bricks, about 20nH I think for that package.  Even if the spikes aren't enough to risk over-voltage for the IGBTs, the rapid current transition at diode snap-off can cause more interference with control circuitry or even MIDI interrupter or keyboard or whatever is close.  (Sounds like you've done a great job of low inductance.   Congratulations on accomplishing something that many builders don't quite understand or value enough!)

It's a bit hard to tell from just the overview scope image you posted, but I did zoom-in and carefully examine phase lead across the burst.  It definitely looked like phase lagged more as the current increased.  There's something changing delay that depends on current.  There are some possibilities that would go the opposite direction, but the only causes I've thought of for increasing lag at increasing current are IGBT switching times and saturation of the phase-lead inductor tuning slug.  May be a cause that hasn't come to mind yet.  That's the advantage of scoping.  (BTW, even if your control circuitry is tied to line, you can trigger on gate-drive by looping an extra wire through the GDT just for scoping.)  You don't necessarily need to scope each stage.  To check IGBTs, just scope gate-emitter, then scope collector-emitter.  Look at the delay from gate to collector at low and high current to see if it changes much.  Also compare gate timing to current or whatever you are using as the common trigger/reference between tests.  If gate and collector track together and both shift relative to current zero-crossings, then back all the way up to the comparitor output or even input.

Concerning DP10013 lead resonance, the Tektronix P5205 is even worse with its longer leads.  (I have one of each.)  At least for the P5205, I've found that the leads when gently twisted together have an impedance of about 200 ohms.  So I made clips for the ends of the leads with 100-ohm series resistors (200 ohms differential that way).  Step response is MUCH better with the 100-ohm series resistors at the clips.  The P5205 exceeds its 100MHz rating, the DP10013 barely meets it.  That's fine for my needs.  100MHz is more than enough, and the shorter leads load signals less.  My DP10013 works fine down to 3.5V, just a bit lower clipping limit voltage.  That allows operation from a single lithium-ion cell.

Just saw the above edits.  Nice job ruling out phase-lead.  Perhaps IGBTs, or perhaps some other issue such as a faulty diode clamping comparitor input voltage swing.  Agree that you should figure out the issue.  Since it is occurring a ~500A, it might be something that gets much worse by 1200A.  For scoping, I gather that this issue occurs the same with 1 enable pulse per second.  No need to worry about cooling air at very low repeat rates.  Drawer access is an issue I can't help with:)

I have seen issues with mis-adjusted phase lead because the scoping CT had too low inductance, so made the scope trace lead actual current.  Don't think that's your issue - you listed the CT cores and they look fine.

[acobaugh]

Here I'm showing two shots taken with phase lead disabled. Ch1 is through the micsig, C to E. Ch2 is with a 1000:1 homemade CT, with unknown skew. This at least shows the relative difference in phase between cycles.
In order:
- last transition before OCD kicks in
- 2nd to last transition before OCD kicks in

[davekni]

That clearly shows the delay change.  I'd suggest next measuring if that delay difference shows up in Vge.

Since your scope has 4 channels, I'd loop a turn or two through the GDT and add a third trace.  Switching of GDT drive (interrupter output) makes a nice reference point for comparing other signals.  Current can work, but GDT drive has faster transitions, so makes timing comparisons easier.  (Keep current, though, as it's zero-crossing point is still needed.)

[acobaugh]

That clearly shows the delay change.  I'd suggest next measuring if that delay difference shows up in Vge.

Since your scope has 4 channels, I'd loop a turn or two through the GDT and add a third trace.  Switching of GDT drive (interrupter output) makes a nice reference point for comparing other signals.  Current can work, but GDT drive has faster transitions, so makes timing comparisons easier.  (Keep current, though, as it's zero-crossing point is still needed.)

That's more or less my plan. Even the transitions that don't show huge spikes, even with phase lead, are happening way late just looking at bridge output. I went back and looked at some scope shots from last year when it was running at ~67kHz, and all of the transitions, with phase lead, were happening on time or just a bit early. Something is off or different, and I can't believe that a lower frequency alone would cause this.

I also designed some gate PCBs to mount the diodes/resistors for the gate drive, but comparing scopes of those and what they replaced doesn't show any difference in timing/phase shift. If anything the new boards and components ring and overshoot just a bit less.

Only thing I can figure is maybe the feedback CT or GDT are introducing a phase shift at this much lower frequency.

We shall see.

[acobaugh]

I found some time to get more scope captures, this time with a few turns through the GDT. I also swapped out the 1nF for a 2.2nF for C33, which had no effect as expected (but at least I feel better about it).

At this point, I'm leaning towards a problem with the FB CT. Magnetics aren't my strong suit, but I think I'm going to try a higher ratio CT to get the feedback current/voltage down. Right now I'm using a cascaded 50:20:1 CT. Cores are both Fair-rite Type 77 (5977003801 and 5977006401):



Scope shots

Ch1: bridge output
Ch2: primary current (from 50:20:1 cascaded CT, matching the ones for FB/OCD)
Ch3: GDT voltage sense winding

Last few transitions, overview, phase lead disabled.:


Last transition, phase lead disabled,  delta_t between zero cross and GDT switch leading edge: +1200ns:


2nd to last transition, phase lead disabled, -500ns:



Last few transitions, overview, phase lead enabled. No huge spikes on any transitions except the last at ~1kA. :


Last transition, Delta between zero cross and GDT switch = +800ns:


2nd to last, -500ns:


3rd to last, -400ns:


4th to last, -800ns:

[davekni]

Great set of scope captures!  My preference is to have phase lead close to the final image, where the bridge output shows the IGBT's anti-parallel diode conducting for ~100ns before current switches.  That demonstrates that the output current is causing the voltage transition rather than the IGBT turning on.  Traces also show that you have the lowest Vbus parasitic inductance of anything I've seen - very nice work!

A bit higher than 1000:1 might be good, but I'd hesitate to change that ratio quite yet.  Given issues at 500A and your design for 1200A or more, how do you know what ratio to use next?  A scope trace of the comparitor input would be great (right side of R2 or top of D1 or D1) if you can clip a scope probe directly to one of those parts.  That's a more sensitive node than most others, so a long wire soldered to it isn't a good idea.

Is the feedback CT burden resistor the normal 51 ohms of UD2.7 (R1)?  What burden resistor are you using for the scope current traces?

I have a new theory as to what could be the issue, and it would show up only at this lower frequency:  I'm speculating that C6 (1uF) may not be large enough, especially if the actual part is on the low-side of tolerance.  At high current, it may have a enough AC voltage swing to slow transitions through forward drop of D1/D2.  Before changing CTs, I'd try a larger cap (or patch another in parallel).  If that fixes the issue, then there's no need to scope the comparitor input.  If this latest guess is correct, a higher CT ratio would help.  However, to get from 400A where it looks reasonable to 1200A would require 3000:1 CT ratio if C6 isn't increased.

Probably unrelated to this issue, and perhaps unimportant, but I noticed that the GDT waveform is developing more droop on the top side later in the burst, but not on the bottom.  UD2.7 has a slight duty-cycle offset (from 50%), which would cause such an artifact if GDT drive were DC-coupled.  Usually the AC-coupling capacitors (C21, C24, C30, and C31) would eliminate any DC current buildup in the GDT.  Are you using those caps?  Are they the standard 2.2uF and 1uF values?

[acobaugh]

Thanks for the comments on the bus design. I figure it's easier to go a little overboard with things like that, rather than try to correct it later by fussing over switching speeds, snubber capacitance, etc.

I had my scope multipliers set wrong for ch1 and ch2. It really was switching 1000-1100A.

Feedback CT is 51ohms as standard. Scope CT is 10 ohms, so 100A ~= 1V.

I don't have any 10uF caps on hand, so I might try winding a higher ratio CT while I wait for yet another mouser shipment.

C[21,24,30,31] are all 2.2uF and 1uF. I'm not too concerned about the droop, as I believe that can be explained by B-field interference with the probing loops of wire and scope probe itself, or at least this phenomenon matches one that Steve Ward wrote about in one of his papers ("Testing and Verification Waveforms of a Small DRSSTC"). The amount of droop seems to be correlated with the amount of primary current. I have scope captures from a year ago where I was probing Vge directly off an isolation transformer, and even up to 500A it's barely noticeable.

Here are some scope captures of R2:




I also, on a hunch, ran the driver by itself fed by a function generator. By varying the input voltage and frequency, we can see the signal at R2 go from sort of square-ish to not:

First two are at low-ish FB voltage, first at ~70khz, second at ~40khz:


Second two are at the highest voltage my generator will put out, first at ~70khz, second at ~40khz:

[davekni]

Nice scope captures again.  Based on those, I think the issue is almost certainly C6 not being large enough.  Your final signal-generator plots are key.  At 1100A and 1000:1, the feedback input will be roughly +-56V (1.1A * 51ohms), 8 times the +-7V from your signal-generator.  If you picture the top/bottom slope to the semi-square wave at R2 increasing 8x, from ~50mV to ~400mV, the intended zero-crossing step will no longer cross zero until part of the next flat slope has occurred.

Perhaps the easiest change that might work with parts you have around:  Increase R2 from 1k to 2k.  That will double the time constant as would increasing C6 from 1uF to 2uF.

A short-term option is to reduce R1, say parallel another 51-ohm resistor.  That's roughly the same as changing CT ratio from 1000:1 to 2000:1.  Only side-effect is doubling phase-lead.  If there's enough adjustment range in L1, you could dial phase-lead back down.  Depends on what's easier, CT rewinding or R1/L1 tweaks.  (You can also reduce phase lead by adding a resistor in parallel with L1.)

Yes, I agree now that the GDT slope appears to be coupling from primary current.  It shows up in R2 scoping too, swamping the intended signal.  For my own curiosity, for the GDT probing, is that channel-3 probe connected to ground as well as one end of the added GDT winding?  If so, the culprit is likely ground-loops back through the scope.  I'd suggest leaving that probe floating except for the ground connection inherent in the scope BNC connection.  If the probe end is already floating, then the current pickup must be through the loop created by scope probe ground lead to the GDT winding.  For such situations, I often use a coax cable instead of a scope probe, soldering connections to the winding with short external loop area.

[acobaugh]

Wound a new 36:30 FB/OCD transformer last night. Made no difference, but I was planning on doing this anyway to get more headroom on the OCD side of things.

Staring at the scope shots and re-reading what you said, I'm convinced the problem is C6. I did notice Fabricio Franzoli's driver also uses a 10uF cap for C6. I did a quick model in circuitjs and I observed the same behavior there. I'll have a 10uF cap here on Wednesday.

[davekni]

30 * 36 is 1080, so only 8% more than initial 20 * 50.  Not surprising that 8% wasn't enough improvement to notice with the phase-lead issue.  I do expect you'll have the solution Wednesday or whenever you get C6 replaced.

[acobaugh]

Great success! C6 is now 10uF. Here it is switching 1100A at 700VDC: