Another QCW Thread

[Coupling]

I apologize because I'm sure this has been discussed in other threads, but I'm struggling to connect the dots.
I've made a system with pretty high coupling and I have achieved some okay spark lengths before my Agilent supply died...
I am struggling with tuning, though. It seems that my primary series capacitance doesn't make a difference. I can short it completely and the results don't seem substantially different.
Without a capacitor or with an oversize dc-block cap, my system locks at 420kHz. Adding a resonant cap doesn't seem to have any noticeable impact on performance. The system tends toward upper pole operation, which I believe only serves to cancel primary leakage inductance, which would have a diminishing effect as coupling increases. The secondary's standalone fres is 400kHz, so I know 420kHz isn't far off from where it should be once coupled with the primary.
I have tried many different capacitor values and I don't see an increase in current draw at any point. I only tried this at low voltage- I don't know if that makes a difference. Also, I expected to use something around the 20nF range, but capacitors of this size just raise the frequency a lot. Perhaps I need more primary turns.

Any direction would be appreciated.

EDIT: Picture attached

[Rafft]

perhaps you could post some pictures of your setup?

EDIT:
no need mention name. but its better to post pictures and more info for completeness.

are you using UD? phase-shifting? buck? minimum bulk cap 6000uF at least.

to my knowledge, qcw starts to have life at more than 150vBus.

what I normaly do is set a baseline/reference freq of the secondary/topload. then tuning my primary/mmc to THAT freq.

qcw "likes" high impedance primary

[Uspring]

The primary winding looks like a lossy inductive load to the bridge. Lossy, because energy is drawn by the secondary coil. The MMC is added in order to cancel out the inductive part of the primary impedance. This has 2 advantages:
1. You can achieve zero current switching
2. You can input much more current for a given input voltage, since the reactive part of the input impedance is cancelled. So you can have both many primary turns and large primary currents, whcih produce a strong field to drive the secondary.
What kind of feedback are you using?

[Coupling]

perhaps you could post some pictures of your setup?

EDIT:
no need mention name. but its better to post pictures and more info for completeness.

are you using UD? phase-shifting? buck? minimum bulk cap 6000uF at least.

to my knowledge, qcw starts to have life at more than 150vBus.

what I normaly do is set a baseline/reference freq of the secondary/topload. then tuning my primary/mmc to THAT freq.

qcw "likes" high impedance primary

I removed the name.
I am using my own custom FPGA controller. It has digital prediction (not quite a PLL but close enough) and it set to a conservative 250ns lead currently and I achieve ZVS. It uses phase shift modulation.
I currently have 3500uF of bus capacitance, but I am planning to move to battery soon. I may add another 3500uF in parallel for the time being, though.
I have 10uF of film right on the bus, with plans to go to 20uF when I upgrade my driver from 50A to 150A. I am keeping an eye on bridge supply transients and they ring in about 40V over the set bus, which isn't great but I can just keep the bus lower until I upgrade.
I have noticed that it doesn't really do anything below 100V! This is in stark contrast with my previous iteration, the ferrite HFXFMR version: https://highvoltageforum.net/index.php?topic=2503.0
The single resonant HFXFMR version created a plume of output starting at 30V bus. This tripped me up.

The primary winding looks like a lossy inductive load to the bridge. Lossy, because energy is drawn by the secondary coil. The MMC is added in order to cancel out the inductive part of the primary impedance. This has 2 advantages:
1. You can achieve zero current switching
2. You can input much more current for a given input voltage, since the reactive part of the input impedance is cancelled. So you can have both many primary turns and large primary currents, whcih produce a strong field to drive the secondary.
What kind of feedback are you using?

1. I expected to see an inductive ramp on the primary waveform, but even with the primary tank cap shorted, it appears to be perfect sinusoid! I am able to easily get ZCS without primary resonance, at least on the non-shifted leg.
2. If there is not much leakage inductance to begin with compared to the magnetizing component due to high coupling, won't the current increase be modest? This was what I was wondering from the start.

I have noticed, however, that without the primary tank capacitor, detuning from streamer loading seems to be much more severe, based on my experiment hanging a wire off the topload.

Anyway, I've got my isolation transformer and variac out, so it's time to get this thing up and running once and for all.

[davekni]

Also, I expected to use something around the 20nF range

Have you interred your geometry into JavaTC to calculate inductances and coupling factor and frequencies?  Or did 20nF come from some other calculation?

The secondary's standalone fres is 400kHz, so I know 420kHz isn't far off from where it should be once coupled with the primary.

If coupling is high, coupled upper pole frequency will increase significantly even with large (or no) primary capacitance and increase even more with matched primary capacitance.

I have tried many different capacitor values and I don't see an increase in current draw at any point.

Looks like long series strings of capacitors.  Total cap ESR (loss) may lower Q too much to see current increase.  Wiring inductance of primary connections reduces effective coupling too.  I'd suggest pairing primary leads through MMC to H-bridge close together to minimize inductance.  In other words, current flowing to one primary lead should be adjacent current flowing from other lead, avoiding large loops for magnetic field.

The system tends toward upper pole operation,

That's great, but unusual.  Upper pole is good for QCW.  However, standard UD2.x drivers usually tend to lower pole unless primary is tuned to higher frequency than secondary.

I'd suggest first running JavaTC, then using those parameters for a simple analog simulation, at least an AC frequency sweep to see pole locations and resulting primary input impedance (H-bridge load impedance) at those frequencies.  Then adjust primary capacitance in simulation.  Try more or fewer primary turns, first in JavaTc, then copied to simulation.  There are several free analog simulators, both on-line and download.  My favorite is LTSpice.

1. I expected to see an inductive ramp on the primary waveform, but even with the primary tank cap shorted, it appears to be perfect sinusoid! I am able to easily get ZCS without primary resonance, at least on the non-shifted leg.

Yes, I saw the same thing on my low-frequency QCW experiment.  Coupling there was very high, 0.9.   As long as secondary Q is high (ie. before arc is long), resonant (sine wave) current will be much higher than the inductive ramp.

I have noticed, however, that without the primary tank capacitor, detuning from streamer loading seems to be much more severe, based on my experiment hanging a wire off the topload.

Yes, you can see that in AC sweep simulation too.  I was originally thinking of no primary cap for my recent QCW (other than perhaps one large cap in case H-bridge duty cycle wasn't exactly 50%).  Simulation showed that a resonant primary helped minimize frequency shift.  Makes sense.  With only one resonant frequency controlled by secondary capacitance, frequency is proportional to capacitance^-0.5.  With primary cap, that capacitance does not change, so combined frequency changes less.

[Coupling]

Also, I expected to use something around the 20nF range

Have you interred your geometry into JavaTC to calculate inductances and coupling factor and frequencies?  Or did 20nF come from some other calculation?

I modeled my system loosely around existing designs that use capacitors in that range. This was a long time ago and I don't remember what tools I used, if any.

The secondary's standalone fres is 400kHz, so I know 420kHz isn't far off from where it should be once coupled with the primary.

If coupling is high, coupled upper pole frequency will increase significantly even with large (or no) primary capacitance and increase even more with matched primary capacitance.

I'm thinking you must be right and the system frequency is much higher than I realize. It looks like it's going to run quite close to 500kHz, yikes! Should be interesting.

I have tried many different capacitor values and I don't see an increase in current draw at any point.

Looks like long series strings of capacitors.  Total cap ESR (loss) may lower Q too much to see current increase.  Wiring inductance of primary connections reduces effective coupling too.  I'd suggest pairing primary leads through MMC to H-bridge close together to minimize inductance.  In other words, current flowing to one primary lead should be adjacent current flowing from other lead, avoiding large loops for magnetic field.

Yes, they are GE 42L and I know they are not ideal, but I have a deadline and they are what I have on hand. I was planning on using 3 parallel banks, but I don't think I have enough to get the value down low enough, so I'll either have to reduce the primary turns or just accept that my capacitors will be short-lived. The capacitors are arranged so that each adjacent device will have current flow in the opposite direction, and I originally was paying more attention to primary lead dressing/interleaving, but I was doing a lot of tuning on the fly and it didn't seem to make that much of a difference. Final design will optimize this.

The system tends toward upper pole operation,

That's great, but unusual.  Upper pole is good for QCW.  However, standard UD2.x drivers usually tend to lower pole unless primary is tuned to higher frequency than secondary.

I noticed that if I add a lot of capacitance in series with the primary, I am occasionally able to excite the lower pole. This only started happening over 0.1uF. I'm using the custom FPGA controller that you have been helping me with (it seems quite stable since you helped me fix my clocks! Thanks!) but it uses basically the same zero cross detecting current feedback as UD, if memory serves.
As a side, note, if I try to target the lower pole by adjusting my frequency limits to exclude the upper one, the system chooses to stick to the frequency upper limit and hard switch instead of exciting the lower pole. Perhaps this is also due to my lossy capacitor bank.

I'd suggest first running JavaTC, then using those parameters for a simple analog simulation, at least an AC frequency sweep to see pole locations and resulting primary input impedance (H-bridge load impedance) at those frequencies.  Then adjust primary capacitance in simulation.  Try more or fewer primary turns, first in JavaTc, then copied to simulation.  There are several free analog simulators, both on-line and download.  My favorite is LTSpice.

I have effectively been doing this on the bench by injecting my signal generator output into my controller's current feedback input. The peaks are not terribly sharp, pointing again to a low Q system. I am seriously starting to think about overnighting some 942s...
I have had some good results tapping different points on my capacitor bank and letting the feedback lock into the frequency, also.

1. I expected to see an inductive ramp on the primary waveform, but even with the primary tank cap shorted, it appears to be perfect sinusoid! I am able to easily get ZCS without primary resonance, at least on the non-shifted leg.

Yes, I saw the same thing on my low-frequency QCW experiment.  Coupling there was very high, 0.9.   As long as secondary Q is high (ie. before arc is long), resonant (sine wave) current will be much higher than the inductive ramp.

I am a bit surprised because I don't think my primary is all that close to the secondary, especially compared to other builds, but operation suggests that it is. I just kind of winged the geometry based on what I saw others doing. I'm not doing anything exotic like you are with your ferrite. But I can see that exciting the primary with the same frequency without the secondary results in very little current.

I have noticed, however, that without the primary tank capacitor, detuning from streamer loading seems to be much more severe, based on my experiment hanging a wire off the topload.

Yes, you can see that in AC sweep simulation too.  I was originally thinking of no primary cap for my recent QCW (other than perhaps one large cap in case H-bridge duty cycle wasn't exactly 50%).  Simulation showed that a resonant primary helped minimize frequency shift.  Makes sense.  With only one resonant frequency controlled by secondary capacitance, frequency is proportional to capacitance^-0.5.  With primary cap, that capacitance does not change, so combined frequency changes less.

If I tune the primary for streamer loading, as is recommended, I suppose for my tightly coupled system that operation will start at the upper pole and essentially ignore the primary cap. The primary will gradually get pulled into resonance as the streamer lowers the system's Fres. This is very odd to me, as I understand DRSSTCs typically operate in the opposite way- you run the primary at its fixed frequency and the secondary pulls itself into resonance as it detunes. Weird.

[Uspring]

2. If there is not much leakage inductance to begin with compared to the magnetizing component due to high coupling, won't the current increase be modest? This was what I was wondering from the start.

When you drive a DRSSTC with a frequency close to one of its poles, its input will look like a serial tank. The resonant frequency is the pole frequency and added in series to it is a resistor, which accounts for all the losses in the system, the most important loss being the arc load. You can then assign an effective primary Q to this tank, which will describe the voltage-current ratios near the pole and its resonance behaviour.

The current increase depends on the effective primary Q. The primary Q is proportional to 1/k^2, is dependent on the secondary Q due to arc loading and also on tuning, in particular the difference between the secondary resonance frequency and the frequency the coil is running at. If you have a large k, a big arc and are in tune, primary Q can become very low. In that case you won't see much of a primary resonance. As said, this serial tank approximation is valid only near the poles, e.g. when you have ZCS on both legs. That's what you probably have at full power.

[Coupling]

Alright, well, I worked up the nerve to get this thing pumping and I managed to get up to 380VDC on the bus without anything blowing up.
I am getting decent streamers but they are barely surpassing the 2 foot mark, which is a bit disappointing. I have some ideas:

1. Better primary tank capacitors. The 850V/0.33u 42L polyester capacitors I'm using are apparently polyester and this may not be a good choice at these frequencies. I will look into replacing them. I had seen some discussion about these on the TCML archives but they must have been referencing much lower frequency systems.

2. Improved primary. I currently have 9 turns with a conical primary. I have seen far more successful systems both with less and with more turns. I already have a lot of coupling as evidenced by the upper pole dominance. Perhaps there is something else wrong with my primary, like incomplete connection to the individual litz strands. I will run some significant DC current through it later and see how much voltage it drops, because milliohm meters are expensive. I am using rather thin wire to jumper a short distance to the primary that did start getting warm by the end of my test. I suppose I should get rid of that.

Not really sure what the best direction to go in is. I have noticed that tuning the primary doesn't actually seem to help that much. There is definitely a point where too little capacitance reduces performance, but I'm having a real hard time finding an optimal value. I know that resonant peaks tend to plateau out as Q decreases, as is the case during streamer loading, so perhaps that is why.

I tried adding 10pF of capacitance to the secondary by way of a string of 100x polypropylene capacitors. This lowered my operating frequency to the 350kHz-ish range but had no discernable effect on pushing more power.

One other observation is that when I start to tune the system to be too high frequency, the primary current does keep rising. I'm guessing that this is due to not having the streamer to de-Q the system. I do wonder where all this power is going...

Anyway, I'm quite jealous seeing everyone else getting 5 ft+. I wish I had some idea of what was going on here for sure, as I cannot push more than 60A with a 380V bus. Hopefully it was just my capacitor material rookie mistake!

Vid:

[Rafft]

try: again

*PRI more L less C

*standalone Pri/mmc Fres higher than standalone sec/top Fres

*phase lead adjustment surely helps

*bigger topload (at the cost of more Iprim)

*at least 6000uF bus cap for starters

Ive set my Vbus limited to 330vdc. its from a flyback converter 16v8 battery input.

I have a total bus cap 6pcs 2200uF 350v, but currently using only 3pcs in parallel. I need to upgrade bus cap construction and a bigger base for coil.

I have set my ocd limit to 200Apk. my build is for "long arcs with less power". during my past experiments, I have peaked it at about 180Apk@300vbus. my end goal was to acheive x10 sec length (done)

bigger bus capacitance does add a few mS of ramp time.

currently upgrading buck. to check between ferrite vs powdered iron in the output LC.

based on your video, problem could posslibly be the MMC and 60Apk is a bit on the low side.

goal should not be 5ft arcs but "x10 the secondary length" 

cheers

[Uspring]

If you have a sine generator, connect it to the input of your coil and look for current peaks. There should be 3, the upper and lower ones the poles. The width of the peaks should give you an indication of primary losses due to the winding or the ESR. Dunno, though, if ESR is power dependent.

[Coupling]

*PRI more L less C

I will probably try this after replacing the capacitors.

*standalone Pri/mmc Fres higher than standalone sec/top Fres

I've been trying every possible tap on my cap string, so I'm kind of already doing this in testing.

*phase lead adjustment surely helps

I'm definitely losing some power transfer due to not being perfectly in phase, but I am reliably getting ZVS. I may be able to reduce the phase lead slightly. This will be #3 on my list!

*bigger topload (at the cost of more Iprim)

I tried adding a secondary MMC, which I think would have a similar effect as adding more of a topload. I did not see any benefit, and the drop in frequency definitely made sparks less straight.

*at least 6000uF bus cap for starters

I am close, maybe around 5000uF. I only see my bus dipping by about 30V by the end of a pulse.

Ive set my Vbus limited to 330vdc. its from a flyback converter 16v8 battery input.

I have a total bus cap 6pcs 2200uF 350v, but currently using only 3pcs in parallel. I need to upgrade bus cap construction and a bigger base for coil.

I have set my ocd limit to 200Apk. my build is for "long arcs with less power". during my past experiments, I have peaked it at about 180Apk@300vbus. my end goal was to acheive x10 sec length (done)

bigger bus capacitance does add a few mS of ramp time.

currently upgrading buck. to check between ferrite vs powdered iron in the output LC.

based on your video, problem could posslibly be the MMC and 60Apk is a bit on the low side.

goal should not be 5ft arcs but "x10 the secondary length" 

I wish I could pull this much current. I'll have to look for a thread with your coil on it so I can see what's so different about your primary.
Well right now I'm x2 secondary length, so I'd be happy with x5.

If you have a sine generator, connect it to the input of your coil and look for current peaks. There should be 3, the upper and lower ones the poles. The width of the peaks should give you an indication of primary losses due to the winding or the ESR. Dunno, though, if ESR is power dependent.

I have done this by feeding my current feedback with my sig gen. I haven't bothered with the lower pole since it seems impossible to excite in my setup, and I can say that the upper pole is... quite flat. So flat that I started this thread asking if it's possible for a system to be so tightly coupled that the primary capacitance doesn't matter. I clearly have some losses somewhere!

Plan of attack:
1. Switch to polypropylene capacitors- 942s are in the mail
2. Inspect primary for possible faulty connections to the multiple insulated strands
3. Add primary turns
4. Reduce phase lead

[davekni]

1. Switch to polypropylene capacitors- 942s are in the mail

Yes, critical first step.  Polyester caps are TERRIBLE at these frequencies.  I suggest ignoring almost all of your previous test results.  Large MMC capacitance (or shorted) looked good because that lowered ESR, not because it is inherently good.

3. Add primary turns

As above, you need to retest after fixing MMC (using polypropylene caps).  If primary current is still too low, reduce primary turns, not increase.  And increase MMC capacitance to keep primary frequency constant.

4. Reduce phase lead

I'm confused.  I though you were using phase-shift modulation for ramping (rather than a buck converter).

There should be 3, the upper and lower ones the poles.

Uspring:  Is that a typo?  Should be 2 peaks, the two poles.

So flat that I started this thread asking if it's possible for a system to be so tightly coupled that the primary capacitance doesn't matter. I clearly have some losses somewhere!

Yes, losses are in polyester MMC.
Also, "tightly coupled" is relative to something.  Standard DRSSTCs are usually lower coupling than you have.  QCW coils are usually high coupling in comparison.  JavaTC would be helpful here.  Looking at your geometry, I'd guess coupling is around 0.4, or perhaps up to 0.45.  That is in the typical range for QCW coils, not particularly high.  Tight coupling causes the two pole frequencies to be farther away from each other, and allows better energy transfer even when primary and secondary frequencies do not match well.

This is very odd to me, as I understand DRSSTCs typically operate in the opposite way- you run the primary at its fixed frequency and the secondary pulls itself into resonance as it detunes. Weird.

Typical non-QCW DRSSTCs ramp primary current quickly.  Initial few cycles look like primary frequency.  As energy couples to secondary, becomes a mix of upper and lower pole frequencies, which beat together causing amplitude modulation.  QCW coils ramp slowly, so excite mostly a single frequency, either lower or upper pole, not both.  Upper pole is generally better for QCW performance.

Good luck with your new MMC and retesting everything else.

[Coupling]

1. Switch to polypropylene capacitors- 942s are in the mail

Yes, critical first step.  Polyester caps are TERRIBLE at these frequencies.  I suggest ignoring almost all of your previous test results.  Large MMC capacitance (or shorted) looked good because that lowered ESR, not because it is inherently good.

I have swapped the capacitors.

3. Add primary turns

As above, you need to retest after fixing MMC (using polypropylene caps).  If primary current is still too low, reduce primary turns, not increase.  And increase MMC capacitance to keep primary frequency constant.

Whelp, see, I did do this, but I just realized now that when I adjusted my ON time to be 5ms for testing purposes, I forgot to also adjust my ramp rate so I was basically seeing the same results as before with the polyester caps, but it was because I was only getting a fraction of the phase shift.... So I went on with increasing my primary, which has led to me needing about 6nF, or possibly less... it also pushed my secondary frequency up to 500kHz, which I don't love. I will definitely have a tappable primary next time :')

4. Reduce phase lead

I'm confused.  I though you were using phase-shift modulation for ramping (rather than a buck converter).

I am. I was referring to the prediction phase lead for ZVS. My voltage leads my current by a noticeable amount.

So flat that I started this thread asking if it's possible for a system to be so tightly coupled that the primary capacitance doesn't matter. I clearly have some losses somewhere!

Yes, losses are in polyester MMC.
Also, "tightly coupled" is relative to something.  Standard DRSSTCs are usually lower coupling than you have.  QCW coils are usually high coupling in comparison.  JavaTC would be helpful here.  Looking at your geometry, I'd guess coupling is around 0.4, or perhaps up to 0.45.  That is in the typical range for QCW coils, not particularly high.  Tight coupling causes the two pole frequencies to be farther away from each other, and allows better energy transfer even when primary and secondary frequencies do not match well.

I imagine this is similar to the LLC converter where there is a resonance at unity gain, where you cancel the leakage inductance, and another at a lower frequency where you resonate with the magnetizing inductance and get gain based on your output resistance. This upper pole operation is similar to the unity gain state, except our output is a high impedance resonator. That means there is no real voltage gain on the primary side with this upper pole operation, interestingly enough. Just your bus voltage * effective transformer ratio * secondary resonator gain.

This is very odd to me, as I understand DRSSTCs typically operate in the opposite way- you run the primary at its fixed frequency and the secondary pulls itself into resonance as it detunes. Weird.

Typical non-QCW DRSSTCs ramp primary current quickly.  Initial few cycles look like primary frequency.  As energy couples to secondary, becomes a mix of upper and lower pole frequencies, which beat together causing amplitude modulation.  QCW coils ramp slowly, so excite mostly a single frequency, either lower or upper pole, not both.  Upper pole is generally better for QCW performance.

Good luck with your new MMC and retesting everything else.

Thank you for your help. I just re-flashed with the correct ramp rate and 17ms on time and I will see what I get. What you are saying makes sense based on my observations. God I hate tuning.

EDIT: UPDATE!

I am now able to hit a target nearly 4 feet away. Thank you everyone. I do have an issue where when I go beyond 340VDC I get erratic behavior from detuning. Increasing tank capacitance doesn't seem to help, although it's possible I don't have enough granularity on my capacitor taps. I also am running a very low primary capacitance of 6nF, so I'm not sure if that has something to do with it.
Maybe I'll add a tuning inductor to the primary. to see if that helps.

Thanks again everyone!

[davekni]

So I went on with increasing my primary, which has led to me needing about 6nF, or possibly less... it also pushed my secondary frequency up to 500kHz, which I don't love. I will definitely have a tappable primary next time :')

Many (likely most) TC designs include tunable primary coil.  All mine are fixed.  With JavaTC and analog simulation, it is possible to get close enough for a fixed primary to work fine.

I am now able to hit a target nearly 4 feet away. Thank you everyone. I do have an issue where when I go beyond 340VDC I get erratic behavior from detuning. Increasing tank capacitance doesn't seem to help, although it's possible I don't have enough granularity on my capacitor taps. I also am running a very low primary capacitance of 6nF, so I'm not sure if that has something to do with it.
Maybe I'll add a tuning inductor to the primary. to see if that helps.

Thank you for sharing success!  Always fun to hear about that.
Erratic behavior may be due to too-fast ramp rate.  At full power, I'm guessing 20ms will be better than 17ms.  That is if you have sufficient energy in bulk cap to extend for 20ms.
Hard to say anything useful relative to 6nF without knowing JavaTC results (inductances and coupling in particular).
Adding external inductance to primary reduces effective coupling factor.  I recommend against adding inductors to primary since high coupling is beneficial for QCW coils.

[Coupling]

So I went on with increasing my primary, which has led to me needing about 6nF, or possibly less... it also pushed my secondary frequency up to 500kHz, which I don't love. I will definitely have a tappable primary next time :')

Many (likely most) TC designs include tunable primary coil.  All mine are fixed.  With JavaTC and analog simulation, it is possible to get close enough for a fixed primary to work fine.

I am now able to hit a target nearly 4 feet away. Thank you everyone. I do have an issue where when I go beyond 340VDC I get erratic behavior from detuning. Increasing tank capacitance doesn't seem to help, although it's possible I don't have enough granularity on my capacitor taps. I also am running a very low primary capacitance of 6nF, so I'm not sure if that has something to do with it.
Maybe I'll add a tuning inductor to the primary. to see if that helps.

Thank you for sharing success!  Always fun to hear about that.
Erratic behavior may be due to too-fast ramp rate.  At full power, I'm guessing 20ms will be better than 17ms.  That is if you have sufficient energy in bulk cap to extend for 20ms.
Hard to say anything useful relative to 6nF without knowing JavaTC results (inductances and coupling in particular).
Adding external inductance to primary reduces effective coupling factor.  I recommend against adding inductors to primary since high coupling is beneficial for QCW coils.

I can see the driver frequency jumping between two values and causing the primary tank current envelope to become jagged when the spark gets erratic. I can see the same thing happening when I detune the coil at lower input voltages. My predictor does a pretty good job at recovering from these changes, and from what I see has been able to correct itself within a half cycle.

I can get good results from about 150V all the way to 340V with the same primary tank capacitance. Unfortunately tapping the next few connections in my series-connected capacitor string does not improve this behavior.

EDIT: UPDATE!

Well it took a lot longer to justify why I thought I had a tuning issue than to test out Dave's idea of increasing on time. Needless to say... it works great with 20ms, albeit with a hit to the rail. Won't be an issue once I'm running on batteries... Hopefully.
8XA peak, reliably hitting the ceiling quite often. Have to stop testing before I get nauseous from the ozone!
Time to wrap this baby up. and see how it performs without a good ground...

[Uspring]

There should be 3, the upper and lower ones the poles.

Uspring:  Is that a typo?  Should be 2 peaks, the two poles.

Yes, there are 3 ZCS frequencies but only 2 current peaks.

[davekni]

I can see the driver frequency jumping between two values and causing the primary tank current envelope to become jagged when the spark gets erratic. I can see the same thing happening when I detune the coil at lower input voltages. My predictor does a pretty good job at recovering from these changes, and from what I see has been able to correct itself within a half cycle.

I'd guess that frequency jumping is something to resolve in FPGA algorithm, especially since it is lasting under 1/2 cycle.

Needless to say... it works great with 20ms, albeit with a hit to the rail. Won't be an issue once I'm running on batteries... Hopefully.

Will your strike rail still be grounded when running on batteries?  If not, where will current go after hitting rail?  Through you?

8XA peak, reliably hitting the ceiling quite often. Have to stop testing before I get nauseous from the ozone!

Sounds like great performance!  Do be careful with ozone.  Human smell sensitivity to ozone varies by 1000:1 from one individual to another.  However, health hazards do not track smell sensitivity.  And continued exposure reduces smell sensitivity.  So smell is not a reliable way to keep yourself safe long-term.

[Coupling]

I can see the driver frequency jumping between two values and causing the primary tank current envelope to become jagged when the spark gets erratic. I can see the same thing happening when I detune the coil at lower input voltages. My predictor does a pretty good job at recovering from these changes, and from what I see has been able to correct itself within a half cycle.

I'd guess that frequency jumping is something to resolve in FPGA algorithm, especially since it is lasting under 1/2 cycle.

The frequency jumping had a beat frequency of around 60kHz and did change with different tap points on the capacitor string. My controller was able to correct itself on the next half cycle after a prediction that was far ahead due to the frequency abruptly lowering. I'm actually quite impressed with it. It is based on an old 4hv thread where Steve Ward described half cycle-based timing, except instead of averaging together the last several cycles to predict, I predict two half cycles at a time using the timing of the previous full cycle, more similar like how a true pll works only on rising waves.

This is gone or at least very rare after taking your advice and increasing the on time slightly with slowing the ramp rate accordingly. You are a legend.

Needless to say... it works great with 20ms, albeit with a hit to the rail. Won't be an issue once I'm running on batteries... Hopefully.

Will your strike rail still be grounded when running on batteries?  If not, where will current go after hitting rail?  Through you?

When I said "hit to the rail" I meant a drop in bus voltage. I don't have a strike rail, but this raises and interesting point. I don't know where the weak link would be. Possibly the isolation barrier between my semiconductor and ground, which would be very disappointing. I think I will ground the negative bus at the bridge, or at least put a film cap there to pass the hf current to ground if there is a strike to the primary.

I'm getting all the secondary base current, and I don't intend to use the system without a full suit of chain mail.

8XA peak, reliably hitting the ceiling quite often. Have to stop testing before I get nauseous from the ozone!

Sounds like great performance!  Do be careful with ozone.  Human smell sensitivity to ozone varies by 1000:1 from one individual to another.  However, health hazards do not track smell sensitivity.  And continued exposure reduces smell sensitivity.  So smell is not a reliable way to keep yourself safe long-term.

I am able to detect it before others. I can also smell it faintly from the static when I slip under my covers when wearing wool pajamas. So I think I am on the more sensitive side, but I will probably start to dull my sensitivity with enough messing around with this.

[davekni]

When I said "hit to the rail" I meant a drop in bus voltage.

Now I understand.  Primary voltage ramp clips (limits) at the end of ramp (duty cycle hits 100%).  So far I've been running that way on my QCW coil, with ~2ms of on time after duty cycle hits 100%.

I think I will ground the negative bus at the bridge, or at least put a film cap there to pass the hf current to ground if there is a strike to the primary.

Yes, I'd recommend that.

I am able to detect it before others. I can also smell it faintly from the static when I slip under my covers when wearing wool pajamas. So I think I am on the more sensitive side, but I will probably start to dull my sensitivity with enough messing around with this.

Wow.  I thought I was on the sensitive side, but you are even more so.  Though I am getting less sensitive as I age (perhaps because of more playing with HV toys).

[Coupling]

When I said "hit to the rail" I meant a drop in bus voltage.

Now I understand.  Primary voltage ramp clips (limits) at the end of ramp (duty cycle hits 100%).  So far I've been running that way on my QCW coil, with ~2ms of on time after duty cycle hits 100%.

I shut down right around the end of the ramp. Maybe I'll play around with dumping some extra juice after I get some heat data on my bridge. Thankfully there's a NTC integrated so this is just a matter of having the discipline to sit down and do it. Maybe in a couple weeks.

I think I will ground the negative bus at the bridge, or at least put a film cap there to pass the hf current to ground if there is a strike to the primary.

Yes, I'd recommend that.

After giving it some thought, grounding the secondary to the negative rail could be bad if a streamer strikes the primary while the high side switch is on. That would put the battery right across the secondary. I will keep the battery floating with a spare 942 film cap from the secondary ground to negative rail to pass any HF from a strike. Another decade-old idea from Steve Ward... actually more.

[davekni]

That would put the battery right across the secondary.

The secondary is relatively high impedance and probably high enough DC resistance that it would not cause any issue being across bus power.  But coupling with a capacitor is fine too.

[Coupling]

That would put the battery right across the secondary.

The secondary is relatively high impedance and probably high enough DC resistance that it would not cause any issue being across bus power.  But coupling with a capacitor is fine too.

I realized this shortly after posting... If I can trust 50kv+ across the secondary then surely 400v would be okay. The primary would still be switching at the same frequency. I'll probably direct ground.