High Voltage Forum

Tesla coils => Dual Resonant Solid State Tesla coils (DRSSTC) => Topic started by: Weston on April 05, 2020, 10:47:07 PM

Title: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on April 05, 2020, 10:47:07 PM
I have been sporadically working on a new QCW tesla coil over the past few months, now with the lockdown I have been making more progress and have decided to make a project thread for it.

The goal is a portable QCW tesla coil with something like 2'-3' long sparks. This is building off my earlier work on my "CuteQCW" and largely shares the same FPGA based controller architecture https://highvoltageforum.net/index.php?topic=587.0

My original plan was to have the coil be mounted on the front of a bike to bring to burning man. Due to storage constraints and the fact I found some cheap electric chainsaws with 120V batteries when looking for a power source for the coil I am now planning on mounting this on the chainsaw frame and having the secondary replace the chainsaw blade.

Due to wanting to mount the entire coil on a bike/chainsaw the electronics are designed to be compact. All of the electronics except for the MMC fit in a 4"x8" footprint, which was what I measured I could fit inside the frame of a bike and still have my legs clear while peddling.

The architecture is a CCM boost converter off a 120V battery pack to supply a bus voltage of ~450V. The power stage is a single SiC MOSFET fullbridge module, F423MR12W1M1B11BOMA1. Its a 1200V 50A module and switches fast enough where I can do phase shift modulation for a full ~20ms pulse and have sufficient thermal margin. Gate drive is provided with isolated DC/DC converters and isolated gate drive ICs, which provides a bit faster switching and allows me finer control over the switching patterns than would be achievable with a GDT.

The github repo is here: https://github.com/westonb/biwheel-coil

Right now I have the boost converter working in constant off time peak current control, which I chose for its simplicity. It also allows me to easily set the maximum current draw from the battery which is the real constraint in charging the bus capacitor. I do not intend for the boost converter and the QCW power stage to run at the same time. They share a single ADC for the boost current mode control and the QCW over current protection.

The QCW driver is based on a digital PLL and seems to be working pretty well with digitally controllable phase lead. Here is a ~1.4ms current ramp during testing, driving a RLC load at low power on the testbench. You can see
 [ Invalid Attachment ]

Here is a photo of the driver electronics during assembly from a few months ago. You can see the bulk energy storage capacitor sits directly over the control electronics to save space.
 [ Invalid Attachment ]

Bringup of the verilog modules for the boost converter and QCW driver are mostly done, right now I am mostly working on integration of everything. All the modules are sequenced by a RISC-V softcore which I can program in C and the coil is controlled by a fiberoptic link. For a portable / handheld coil I am going to have to make some sort of control board. 

In the next two weeks or so I hope to have integration of the verilog modules done and then I can move on to some more testing. Right now I am using my MMC from my old coil and a roller inductor to emulate the tesla coil. It might be hard to test with the secondary due to the lockdown but I am sure I can figure something out. I have only brought the power stage up to ~60V but all the waveforms seem to be good. I am really hoping I don't blow up anything because it might be a bit difficult to assemble a second board during lockdown.
 [ Invalid Attachment ]


Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: davekni on April 06, 2020, 12:43:50 AM
I like the overlapping copper sheets for your bulk capacitor!  That's a great low-inductance design, better than side-by-side bus bars.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 02, 2020, 10:36:44 AM
It took a week longer than I planned but I finished all the verilog. I ended up not moving everything to a AXI4-lite bus because of the added complexity.

Due to timing constraints the RISC-V softcore is running ay 80MHz while the control state machines and all the other peripherals run at 240MHz. I have realized I never should be buying the lower speed grade FPGAs, the extra time spent trying to resolve possible timing issues (and extra time having the synthesis tool trying to route the FPGA with worse timing constraints) is not worth the marginal dollars saved.

Today I got my first sparks with the old secondary / MMC from my previous QCW coil. Small sparks, but sparks never the less! Attached photo did not capture the spark at the full size, but its not that much bigger. A fair amount more work until I am pumping out 3'+ sparks like the previous coil, but progress never the less.

I am having an issue with EMI crashing the FPGA. I am not sure what exactly is causing it but I suspect its related to the programmer remaining connected. I am going to remove the un-needed test leads I had for debug and load the bitstream on the flash memory later this weekend and see if that helps. For a future design I think I am going to try to put all the control electronics under a shield can.

On the theory side of things, I am seeing almost a constant current as I ramp my phase shift after the coil has breakout, its almost acting as a current source. I seem to remember discussion about this before? It also matches what I saw with my previous coil.  It does not match what Loneoceans saw in his QCW coil https://www.loneoceans.com/labs/qcw15/ https://www.loneoceans.com/labs/qcw15/23MayFirstLight.png . That coil used a buck converter modulator, but that should not impact it.


 [ Invalid Attachment ]


Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on May 02, 2020, 04:31:23 PM
A very neat little coil  ;D
The shape of the primary current envelope depends much on the tuning. There are 2 opposing effects: If the driving frequency is very different from the secondary resonance frequency, the lowering of secondary Q due to the larger arc loading later in the burst will widen the resonance curve and increase the power transfer to the secondary. That will keep primary current low.
If driving frequency and secondary res frequency are well matched, then the dominant effect of larger arc loading will be the reduction of secondary current. That will decrease power transfer to the secondary. Both effects are captured in the equation:

Qpri = (Qsec/k^2) * (1 - f^2/fsec^2)^2 + 1/(k^2 * Qsec)

Qpri describes, how far primary current will rise for a given bridge input voltage. Depending on whether f is near fsec or not, the primary Q will either rise or drop with dropping Qsec. I suspect, that you're running the coil a bit distant from fsec (=secondary res frequency). How does your feedback work? If I understand this right, you are hard switching the primary for power modulation. The current zero crossings, which define the pole frequencies, might be modified.

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 02, 2020, 11:34:04 PM
I ran some more tests today. I added some shielding and loaded the bitstream on to the configuration flash so I can disconnect the jtag programmer. No more crashes! With that I have brought things up to 90V input and got a somewhat bigger spark. Its hard to take these spark pictures, I need to set up a camera tripod...



90V is the limit of my two lab power supplies in series. For higher power I am going to have to enable the boost converter on the input. This will require writing a command parser so I can set the bus voltage over the fiber optic serial link. Eventually I am going to be running the entire coil of a 120V lithium-ion chainsaw battery.

Beyond that, everything seems to be running decently smooth. Here is a scope capture of one of the pulses with a 90V bus voltage:



I am not getting a monotonic ramp current because I a slightly oscillating around the resonant point. A lower Q with larger sparks / higher power may dampen this, otherwise I am going to have to tweak my loop filter. The high peak current at the beginning before breakout is also a bit concerning. I probably have the thermal margin for it, but I should be able to eliminate this by starting sufficiently above resonance and then backing down once things stabilize.

Looking at a close up of the switching waveforms it seems like I might want to reduce my phase lead by ~100ns or so, better to switch too early than too late though. The ringing looks a bit ugly, but I think a lot of that is due to my probing technique, a switching transition on one node should not induce that much ringing on the other node.



My peak primary current is already at ~32 amps for a 90V bus voltage. It seems to be rising very sub-linearly with bus voltage (Almost the same at 60V, ~22A at 30V) but it looks like I might need to reduce my MMC capacitance and use a higher turn count primary for the final primary/secondary configuration. I can push up to a 500V bus voltage and want to keep peak primary current under 60A or so. The SiC brick I am using (datasheet:  https://www.infineon.com/dgdl/Infineon-F4-23MR12W1M1_B11-DS-v02_00-EN.pdf?fileId=5546d462689a790c01690e9d9fb63802  (https://www.infineon.com/dgdl/Infineon-F4-23MR12W1M1_B11-DS-v02_00-EN.pdf?fileId=5546d462689a790c01690e9d9fb63802)) has very low switching loss so I should have a large thermal margin, but I really don't want to blow this up. BOM cost was high and its all integrated, so one failure and the whole PCB is toast.

Uspring:
Thanks for providing the explanation. Is "f" the primary resonant frequency? I am running on the upper pole with the secondary tuned above the primary, so based on that formula spark loading should pull things into better tune and reduce the primary current. This primary / secondary + MMC is from my previous QCW coil which was detuned a fair amount to account for loading from ~3' sparks.

My driver uses a digital PLL which allows me to lock on to the upper pole when primary is tuned below the secondary if I supply a small number of fixed frequency starting cycles to energize the correct pole.

One side of the bridge switches on the zero crossings of the tank current with an adjustable phase lead. The other side switches with an adjustable phase offset (aligned such that I get inductive loading on the hard switched leg). This configuration does mean that the effective phase of the applied voltage and the current varies as I sweep the phase offset.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on May 03, 2020, 05:29:44 PM
I am sorry about the missing explanation of the equations terms. f is the bridge frequency and fres the one of secondary resonance.
Since you are running at low power, Qsec is probably large, making the first term of the sum

Qpri = (Qsec/k^2) * (1 - f^2/fsec^2)^2 + 1/(k^2 * Qsec)

dominant even if tuning is good, i.e. f near fsec. As you increase bus voltage, the increase in arc power will lower Qsec and possibly keep the primary current at bay.

Another maybe minor issue is the effective phase difference between primary voltage and current you mention. Since you are on the inductive side, the bridge frequency f might be a bit higher than the upper pole frequency. The difference f - fupperpole is probably small due to the high Qpri, but it increases f/fsec since f > fupperpole > fsec. That causes somewhat high primary currents than expected, when the bridges are phased for low power output. At max power, i.e. 180 degrees between your bridge outputs you'll be back to f=fupperpole, though.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 04, 2020, 10:10:57 AM
Ah, and for a normal coil the bridge frequency is the frequency of the dominant pole. I have been delaying but I am going to have to go through and derive some of the equations myself. And also possibly figure out a more optimal ramp pattern than a linear phase ramp. The relation between power delivered and phase is not linear.

If Lone Ocean's write up is accurate, the waveform of his linked is with a different tuning, most likely  f_pri > f_sec, driving the upper pole. It makes sense that the difference in tuning would account for this behavior, I am just surprised to see such effects in my own waveforms with such small sparks. Perhaps I am going to be able to stay under ~60A or so at full bus voltage. Its interesting to think about how a different driver can allow you to utilize your switches a lot more effectively. Anyone have any estimates of loaded secondary Q? I might try to calculate the difference in switch utilization you get for different tuning conditions.


In project log land, I spent a long time this weekend battling with software libraries and trying to build toolchains from source. The FPGA has everything sequenced by a RISC-V softcore (its open source! https://github.com/cliffordwolf/picorv32 ) and all the tesla coil control stuff connected as memory mapped peripherals. This is how I structured my previous tesla coil driver and also a setup I have been using for some of my research projects. In the past, library support has been pretty much non-existent, which makes basic stuff that would be accomplished by the C standard library pretty painful. Now that I want to have a full serial command interface I need string parsing, which is implemented in the C standard libraries.

My project is now set up to use picolibc ( https://github.com/keith-packard/picolibc ) which provides a light weight implementation of the C standard library for embedded devices. To support an system for basic input/output commands you only need to provide basic putchar and getchar functions, which is nice. Getting it to build was another issue and took up most of my day. I had to update my compiler to a more recent version of the GCC RISC-V toolchain to build the library. I was having issues getting the library to build with the riscv-gnu-toolchain that I was building from source. It turned out to be something related to the compiler build process producing a compiler with issues and switching to the toolchain provided by SiFive worked. Apparently RISC-V support is now in the mainline GCC releases, so I might investigate that in the future.

Now that I have basic library support for the processor it should not be that difficult to write a serial parser and control the boost converter over the fiber link to do higher power bringup. I am hoping to have that done and do higher power tests by next weekend, I might have to order an extension cord to do testing in courtyard of my apartment complex, the current setup is in my apartment bedroom and its pretty crowded.

No project update is complete without a photo, so here is a photo of my current test setup. Its a bit tight, this photo guest stars my bathroom towel and the kitchen chair:

 [ Invalid Attachment ]

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on May 04, 2020, 05:18:46 PM
It looks like Loneoceans had fpri > fsec on his first version, but later fpri < fsec on the "Fat toroid" model. That can, as you say, account for the different behaviour. Loneoceans has quite long arcs as compared to the toroid size and that implies markedly changing tuning conditions during the burst. If you want to keep primary currents small it is important, that secondary resonance stays above primary resonance. If this is not the case the upper pole frequency won't follow the secondary res as it is going down. A short illustration: (fupperpole is also the running frequency f)

---flowerpole----fpri-----fsec----fupperpole----      ; initially
---flowerpole----fsec-----fpri----fupperpole----      ; medium power
---flowerpole----fsec-------------fpri----fupperpole----      ; high power

Better is:

---flowerpole----fpri---------------fsec----fupperpole----      ; initially
---flowerpole----fpri-------fsec----fupperpole----      ; medium power
---flowerpole----fpri-fsec----fupperpole----      ; high power

It is a bit weird, but in upper pole operation, with the primary tuned low, it does not really matter at which frequency the primary tank resonates, as long as it is below the secondary fres.

My best guess for Qsec comes from Wards top voltage measurements with his sword sparc QCWs. Typical voltages were around 70 kV peak and only weakly dependent on length of the arc. You can derive a ballpark arc loading resistance by using your power level and this voltage value.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 10, 2020, 09:32:43 AM
Time for a project update!

I am planning on deriving the tesla coil transfer function and seeing how that behaves with a load that clamps the voltage, which should help me optimize my new secondary / primary. However, I have been putting that off as I can always work on theory while waiting for new parts / PCBs if I blow something up  ::)

In order of progress, I:

Changed the picorv32 soft core configuration / makefile from RV32I to RV32IM instruction set (soft core now supports multiply and divide instructions!)

Wrote a serial parser that can support arbitrary commands and values, its loosely based on the SCPI command syntax set used for test equipment. This allows me to call functions / set register values for things like burst length and boost converter target voltage from my laptop, which is really helpful for bringup, it allows me to test things without having to generate a whole new FPGA bitstream. This it the first time I have had arbitrary bidirectional communication with the FPGA, exciting! I should be able to backport this setup to my PhD research projects which will be helpful in the future.

Brought the boost converter up to 300V output and made some tweaks. The boost converter runs in constant of time peak current control because that was the quickest thing to write a controller for in verilog. I fixed a minor issue with my minimum on time being too low, causing runt pulses. Related to this and also fixed was issue with inductor current exceeding my set point during start up when Vin is close to Vout. This was caused by the minimum on time and fixed off time, I fixed it by keeping the FET off until the current goes below the limit.

With constant off time I enter discontinuous mode at higher conversion ratios, shown here (yellow gate drive, orange inductor current, green switch node voltage, blue output voltage):
 [ Invalid Attachment ]

I might want to keep things at critical conduction mode to reduce switching loss, but this constant off time control is leading to poor utilization of my semiconductors / inductor and causing more ripple on the DC input. Its not that critical now as I am not trying to achieve high BPS at this point in the bring up, but at some point I want to rewrite the controller to operate in CrCM or CCM. The inductor current is sampled with the same ADC I use for over current protection (80MSPS) so I have a lot of control flexibility.

This bringup also allowed me to verify the switching waveforms of the boost conveter. I plan a maximum DC bus voltage of 500V and the boost converter uses a 650V super junction FET. The To-247 leads are at their maximum length due to the standoff from the heatsink imposed by the SiC MOSFET module so I was a bit worried about stray inductance and voltage spikes during switching. Scoping things out shows minimal ringing during turn off (Yellow: gate, orange: inductor current, green: drain):

 [ Invalid Attachment ]

I also brought up the coil to 300V DC bus voltage. Large sparks are damping the secondary Q and eliminating oscillation of the PLL loop filter, but the discrete nature of the oscillator is still evident. I have a 240MHz clock frequency for the PLL module on the FPGA which gives me ~ 660Hz resolution at 400KHz. Looking at my current waveforms, it seems this is leading to quantization error which is showing up as slight steps in the current envelope. These are small so it does not seem like it would be worth spending effort to rectify, but its interesting to observe. I should be able to get a higher frequency resolution with dithering.

 [ Invalid Attachment ]

In bringing the coil up to 300V I have new longest sparks from the coil! I apologize for the blurry photo, I need to set up a tripod. This exact spark / photo also marks the point at which I realized that any higher power testing will need to be done outdoors:

 [ Invalid Attachment ]

Lastly, I also tested the coil with the battery for the first time, with everything powered directly from the 120V power tool battery. given the high current rating of the battery, I was a bit nervous plugging it in, but everything seems to run fine! This should make outdoor testing easier, I only need to run a power cord for grounding and the scope.

Now that the power electronics have mostly been brought up its time to focus more on repurposing the chainsaw enclosure to house the electronics, making a stand along interrupter type controller for the coil, and doing enough theory work to optimize a new primary / secondary for the coil.   

Also, as a question to the readers, does anyone have advice on grounding for a portable (hand held) coil? I have been connecting the secondary into mains ground and also have a few square feet of foil on the chair right now to act as a sort of counterpoise. I had the mains ground get disconnected during testing and the over current detection would trip (55A right now) and I would get a very small spark. I assume I will have more capacitance that a few sheets of foil and I can connect the ground to my shoes to hopefully get a ground return to whatever I am standing on, but it has me a bit worried. 
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 16, 2020, 09:16:15 AM
I have some more testing planned this weekend (which will test predictions of my recent modeling) so I figured I should make a pre-update.

This week I took apart one of the electric chainsaws I have and decided that it was going to be too much effort to modify the existing housing to fit my driver and the coil itself. I decided it would be easiest to 3d print one and bought the biggest/cheapest 3d printer I could find, the ANYCUBIC Chiron. It has a 400x400x450mm bed size so I should be able to print the enclosure in the minimal number of pieces (the original chainsaw plastic housing is ~18" long). I am also upgrading the hotend to the E3D volcano to get a faster extrusion rate and not have to wait forever for the prints to finish. I think it would be cool to print a housing out of translucent plastic and insert some LED strips. I am planning on having a ESP32 based controller board internal to the case with its own battery and connected over the fiber link (its really the only interface I exposed....) and some switches + a display for control. It should be pretty cool looking! I just need to decide what plasma/energy weapon from scifi/video games I want to base the enclosure design off of  ;D

My main achievement this week was spending a lot of time  working on tesla coil modeling. I really want to optimize the utilization of my power electronics and achieve maximum primary current at maximum bus voltage. Given that I only want to make one new primary/secondary, this is going to require an accurate model of the primary impedance under spark loading. My old primary / secondary used on my cuteQCW coil seems to have a good impedance match for keeping tank currents at ~ 50A or so, but that was really just luck. Further complicating this, I am considering using a ferrite core in my coil which javaTC can not model. I have been testing my coil with the primary/secondary from my cuteQCW system which serves as my reference point for testing my modeling. The modeling is split into two parts: accurately modeling the primary / secondary and accurately modeling spark loading.

For modeling the primary / secondary itself I am using FEMM to model the magnetics. I have wrote a script which creates my primary, secondary, and topload in FEMM using the octave scripting interface (the UI for manual entry is a pain)  https://github.com/westonb/biwheel-coil/blob/master/modeling/cuteqcw_secondary.m . I can directly verify the inductance of my primary generated from this model, can sort of verify the inductance of the secondary by measuring the resonant frequency, and measure the coupling between the two by looking at the measured frequency difference between the upper and lower poles. For the capacitance of the secondary I am using the Cl-DAE formula (which is derived from the medhurst self capacitance formula) from this very excellent paper: http://g3ynh.info/zdocs/magnetics/appendix/self_res/self-res.pdf and for the capacitance of the topload I am using the deep fried neon toroid capacitance calculator (I really need to find out where this formula comes from) http://deepfriedneon.com/tesla_f_calctoroid.html .

This is only for N=1, but so far my model seems to be more accurate than JavaTC. JavaTC predicts my secondary resonant frequency to be 301KHz, my modeling predicts it to be 285KHz and my measurements put the resonant frequency at 285.1KHz. That match is way closer than the uncertainty of the empirical  formulas for capacitance but hopefully I did not just get super lucky in this one case and its actually accurate! Java TC predicts my coupling at 0.433, FEMM predicts the coupling as 0.426 while fitting the rest of the modeled system to my measurements gives a coupling of 0.47. I believe this discrepancy is due to the self resonant effects of the secondary. When running close to a waveguide resonant mode the secondary winding current is non uniform. In the quarter wavelength resonant mode the current at the top drops to zero. Based on this, windings closer to the base are going to have a higher effective coupling factor. No easy way to incorporate this into my model but the effect should reduce with added capacitance from  larger sparks / topload which moves the secondary away from the quarter wavelength resonant mode.

One interesting thing from my modeling is that I can model the impact of the topload acting as a shorted turn. At least for my system (which has a relatively close topload) the topload has a relatively minor impact on the impedance and loss of the secondary. At 380KHz the secondary with the topload has an impedance of 148.569+I*52889.5 Ohms (thats a 22.15mH inductance) and the secondary without the topload has an impedance of  146.103+I*54592 Ohms (thats a  22.9mH inductance). In reality the effect should be even less due to the previously mentioned non-uniform current distribution. Here is a  plot from the FEMM model (which is axisymetric, hence the missing half) showing how the topload distorts the magnetic field when the secondary is excited with a current of 1A:



Once I had a decent model of the static tesla coil system I moved on to the modeling of the impedance of the system under spark discharge. Based on my previous measurements of relatively constant tank current and other peoples reports of the streamers clamping the secondary voltage to a ~ constant value I hypothesize that the primary + secondary acts as an immittance converter, which converts a voltage on one port (the secondary) to a current on the other port (the primary). As the spark grows and adds capacitance but the voltage (in theory) stays relatively constant the characteristic impedance of the immittance converter changes and leads to a change in primary current. With my current spark length it seems that Cspark << Ccoil+Ctopload so the characteristic impedance stays relatively constant, hence the constant current.

I tried modeling my coil with the spark model provided by Uspring here https://highvoltageforum.net/index.php?topic=1073.msg7715#msg7715 but it leads to signifigantly higher primary currents than I have measured (~70A vs ~43A measured). Based on the start of the pulse it seems that the initial topload clamping voltage (~38kv) leads to a correct initial current, but the streamer model adds too much capacitance / the topload voltage rises too much, leading to an increased current beyond what I observe. I made my own simple model that uses a voltage source to clamp the topload voltage to a constant value of ~35kv. For streamer capacitance I measured my coils spark loading by looking at the frequency difference between the start and end of a burst and calculating the change in capacitance that would be required to achieve that (~2pF for a ~2' spark). I ramp this capacitance from zero the sqrt of the modulation ramp. Its a predictive model of spark loading, not reactive, but it seems to give decent results. Here is the current envelope I showed previously for a 300V bus voltage and here is the model, it seems to be pretty close.





This weekend I am planning to test my coil outside for the first time and run it up to a bus voltage of ~450V to verify the power electronics work at full power and  collect more data to validate my (simple) spark model. I also modified the verilog for the controller of my boost converter to run it in something close to critical conduction mode and lead to faster charging of the bus capacitor. Once I have more data on spark impedance at this frequency I can work on designing a more compact primary / secondary for the chainsaw coil along with an enclosure.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 19, 2020, 07:55:54 AM
This week I validated my boost convert up to 500V and tested the coil outside for the first time.

In validating the boost converter I found a minor (but potentially destructive) issue where the current sense was mux'ed away when the last boost switching cycle was still occurring, leading to a loss of current limiting. That was a relatively simple fix. I also added a switching threshold for turning the FET back on after inductor current drops to zero to keep the system in something like boundary conduction mode, which will allow for faster charging of the bus capacitor for a given peak inductor current.

After working on the boost converter I tested out a "burst" QCW mode where I had multiple pulses in quick succession and ended up destroying my nice DC load and almost destroyed the driver  >:( . The ADC on the FPGA that I was using to sense the bus voltage cut out and the boost converter did not shut down when it should have, eventually suppling something greater than 500V and dumping 500+j into the DC load, completely destroying it. It seems all the pass transistors in the load are blown and the event blew the DC load fuse and my breaker (which I am quite confused about). The DC load seems like a total write-off. On the bright side it protected my driver from a similar fate, which would have only been a little less expensive and a lot more difficult to replace. Here is the most impressively blown FET, but all 8 in the load suffered similar fates. My coil also killed a USB hub so it now has a 2/0 K/D ratio  ::)



I believe this failure was caused by leaving he JTAG programmer connected to the FPGA during testing. It was previously causing crashes when left connected at higher powers but I had not previously had issues at lower powers. I believe the previous crashes were due to the programmer cable picking up EMI which was interested as JTAG commands and causing general weirdness / resetting the FPGA (the FPGA is protected with a level shifter so I don't think it was EMI directly causing problems). The XADC on the FPGA I am using is also directly accessible from JTAG. I think an invalid JTAG command may have reset / halted the XADC and lead to the failure to regulate the boost converter output voltage. After this incident I have added some sanity checks on the ADC output, disabled JTAG access to the XADC once the bitstream is loaded, and I will no longer be testing anything with the programmer left connected.

Today I tested outside for the first time! The coil was running off the battery but I had to run an extension cord for my scope and some supplementary grounding. Results were decent, it looks like my longest spark is ~3' so far. I ran these tests at a ~10ms burst length and a 450V bus voltage. The peak tank current is 80A, which is a bit higher than I would like. Prior to testing I reran the switching loss figures and I should be able to run up to 80A peak tank current but it makes me a bit nervous, especially if I want to go up to 20ms. Also, it seems my OCD is off by ~ 10%. I swapped out the current transformer and need to adjust the gain.

Here are photos of a decent spark and the tank current envelope:

 




I also recorded some 20M point data sets I can look at to determine how much the coil detunes during operation and use that data to test the spark models.

The sparks are more branched than I would like, I think that is due to some combination of ramp rate / pulse time and weather. I am going to mess with ramp rate and go up to 15ms+ in the future.

Over all, this testing was a success! Limited by the rating of the DC bus capacitor, I can bring the coil up to 550V but I am limited by peak tank current right now. This new data will help me better design a secondary / primary for the final chainsaw tesla coil.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: SteveN87 on May 19, 2020, 02:08:20 PM
Quote
I believe the previous crashes were due to the programmer cable picking up EMI...

I've seen a case where some FPGA-containing equipment failed EMC radiated immunity at 3V/m due to a 3.3V rail being gradually "pumped up" by a parasitic charge pump formed by a cable (connected at the equipment end only) and the external (and presumably internal) clamp diodes of an interface IC. A TVS on the 3.3V rail solved it.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on May 27, 2020, 06:21:39 PM
The FEMM simulation of your TC is pretty. JavaTC takes some shortcuts to avoid excessive computer time for its calculations. I believe they use a cylindrical symmetric geometry, basically a collection of concentric rings, which are either stacked on top of each other for e.g. a secondary or around each other such as in a pancake primary. Then all self and mutual inductances are obtained. So it is a magnetostatic model, which does not take into account dynamical, i.e. non uniform current distributions in the secondary. In a similar way capacitances are calculated. There is a possibility to look at the secondary current distribution with JavaTC if you run the voltage and current distribution option. I think, that you are correct about the underestimation of k due to the non uniform secondary current.

My arc model indeed overestimates primary current. The problem is somewhat aggravated by the positive feedback of the overestimation: Too much arc capacitance leads to higher primary currents, which implies more input power, which lengthens the arc, which again means more arc capacitance. Modelling the arc instead by a rule, which makes its capacitance proportional to its length looks plausible. In a calculation for steady state arcs, i.e. ones which don't grow too fast, I get exactly this proportionality. https://www.pupman.com/listarchives/2012/Oct/msg00125.html .That model doesn't clamp secondary voltage, but it shows a steep rise of power with increasing voltage, e.g. V^4.

So assuming a clamped output voltage is close to reality. I do have a problem, though, with the clamp voltage you chose. It implies at the 8.5kW power level an arc load resistance of about 80k. I don't know the exact arc capacitance you assumed there, but I guess it to be around 2p (Arc res and cap are thought to be parallel here). That is in strong conflict to Hydrons and my arc measurements,which indicate a roughly 45 to 60 degrees phase shift between arc voltage and current. The arc model quoted above also has voltage-current phases in that range.
In the diagram below the input resistance of a loaded TC is shown. The blue curve shows the real and the red curve its imaginary part.



The operating frequency for zero current switching is the one where the red curve crosses zero. Strictly that is for a phase shifted bridge only the case for max power. I've tried to get the correct arc loading parameters while keeping the phase shift constraint (45 degrees) but wasn't able to obtain them since the calculated input resistance was always smaller than 6 ohms. The actual one, derived from 300V (square wave) 43A input, is 9ohms.
Can you provide some frequency data from your runs? That would possibly help.

Wrt the too large currents: I believe that tuning your primary lower will reduce the currents. That will also reduce the operating frequency. If you don't want that, you'll need to increase secondary res f. For both measures you'll have to keep an eye to stay at the upper pole. That won't be anymore the preferred frequency. Possibly also current draw and branches are related, i.e. branching arcs might have more capacitance for a given power. Have you observed anything like this?

Congratulations on your recent runs. The arcs look nice. Possibly longer bursts will avoid branchings.

Here's a Maxima script to calculate the input impedance for a TC. The above diagram was made with that:

f(w):=1/(%i*w*C1)+%i*w*L1-%i*k^2*w^2*L1*L2*(%i/R2-w*C2)/(1+%i*w*L2/R2-w^2*L2*C2) $
wxplot2d([realpart(f(2*%pi*w)),imagpart(f(2*%pi*w))],[w,230000,400000]),k=0.426,R1=0.25,R2=220000,C1=9.9e-9,C2=16e-12,L1=33.8e-6,L2=0.0221;
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 29, 2020, 10:50:36 AM
Quote
I believe the previous crashes were due to the programmer cable picking up EMI...

I've seen a case where some FPGA-containing equipment failed EMC radiated immunity at 3V/m due to a 3.3V rail being gradually "pumped up" by a parasitic charge pump formed by a cable (connected at the equipment end only) and the external (and presumably internal) clamp diodes of an interface IC. A TVS on the 3.3V rail solved it.

Thats an interesting failure case, bit I dont think its happening here because all my rails are pretty low impedance. Among other things, the 3V3 rail has the fiber optic transmitter on it which always consumes at least a few mA. I will have to keep this in mind for the interface board I am going to make for this though.


Hi Uspring,

On my most recent run at 450V bus voltage that those waveforms are from, I started at ~390KHz and ended at 330KHz. If you want to look at my recorded current waveform, I have a 20 MSample scope capture here: https://www.dropbox.com/sh/167s5kcb8i3feye/AADXuA1yNu2IvK1kywHbHnOAa?dl=0 . Based on my other measured and modeled parameters, this equates to ~ 7pF of arc capacitance. Using that value of capacitance for my previous spark model gives a pretty close match between the measured waveform and simulation. My secondary coil form is 1' long, so using that as a reference point it seems I am getting up to 3' sparks in my last run. Despite getting a sizable variation in spark length / amount of branching between pulses, all the pulses seemed to have a current envelope within ~10% of each other.

Over the last few days I worked on refining my frequency domain simulations of the resonator and deriving a transfer function for the system. Your observation that increasing the primary capacitance would decrease my primary current is correct. I was a bit surprised by that, it runs counter to my coiling intuition, but it makes sense looking at the model. I have been working on tesla coils since before I had any formal EE education, so I guess there is a fair amount of stuff to unlearn.

I have been putting off designing a new secondary + primary + MMC for the final coil until I have all my modeling down. Other than matching the system impedance to my inverter capabilities I have been thinking about figures of merit to design the resonator and I have realized that the current delivered to the secondary vs the current circulating in the primary inductance is probably important. I have backed out an approximate spark resistance from my last run and have implemented a transformer model with dependent sources that allows me to directly measure the current flowing into the magnetizing inductance.

 [ Invalid Attachment ]

Right now my model suffers from the fact I am using a fixed spark impedance for the frequency sweeps, so it does not model any feedback effects due to spark growth, but it has provided some interesting observations. One observation is that the minimum system impedance (and secondary voltage for the fixed spark impedance) does not occur at a 0 degree phase shift but actually occurs at a ~ 20 degree inductive phase shift. I am going to have to poke around more with the transfer function more to figure this out, but it seems to be in part related to the primary and secondary resistance. Removing those gets it down to 10 degrees, but I still cant get the current maximum to coincide with a 0 degree phase shift in simulation. I double checked with my old simulation using the more standard coupled inductors and its observable there too. My inverter can already hard switch at max current so it might be worth exploring this.  (current flowing into the system in blue, current being coupled into the secondary in green)

 [ Invalid Attachment ]

The second observation is the proportions between the magnetizing current and the current being coupled to the secondary. For my current tuning (and the next tuning i am going to test) the current is at least 10dB less than the current coupled into the primary, worrying about it is likely an optimization of diminishing returns. However, when I change the tuning in simulation (16.7nF primary capacitor) such that the lower pole is dominant, the magnetizing current is larger than the current that is coupling into the secondary. I know that people have gotten better results with upper pole tuning, but the disparity in simulation seems massive compared to reported differences and makes me somewhat question my results. There may be some feedback effect with spark loading that I am missing. (turquoise is current into the system, green is current circulating in the magnetizing inductance, blue is current coupled into the secondary)

 [ Invalid Attachment ]

My next step is to explore the impact of altering the secondary impedance. I am considering operating at a lower secondary impedance and using a secondary MMC to limit detuning. Based on some old posts on 4hv I was going through, this may also reduce the current circulating in the primary magnitizing inductance relative to the secondary and allow me to more fully utilize my inverter, but it seems I may already be getting diminishing returns with that. A secondary MMC should also allow for a smaller topload, which is nice for aesthetic reasons, and reduce displacement currents compared to a larger topload, which is nice for a hand held coil. I should be able to fit ~50 SMD 1812 ceramic capacitors with isolation routing between them. Given the selection of NPO ceramic capacitors this would be ~100kV voltage rating and up to ~15pF if needed.

This weekend I am going to assemble another 3nF of MMC and test the coil outside again to verify the new tuning reduces primary currents. If the primary currents are lower I might bring the coil up to a burst length of 20ms and/or a bus voltage of 500V (my bus cap is rated for 550V). 
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on May 30, 2020, 05:25:45 PM
Hi Weston,

I find it interesting and counterintuitive, that spark length and branching don't seem to effect the current envelopes. Something to think about.

The minimum input impedance is indeed not at the 0 degree phase shift frequency. You can see this also in the diagram of my previous post. Minimum impedance is where the sum of squares of its real and imaginary part are minimal. That's a bit above the 0 degree frequency.

I don't think, you have to worry about magnetizing currents. There is no power ever going back into the bridge the way you set it up. Either you have a voltage in phase with the current, supplying power to the coil, or the bridge voltage is 0, which means zero power transfer between bridge and primary.

Something seems to be wrong with your circuit above. It seems to behave differently compared to its coupled transformer version.

Regarding possible figures of merit: I think, the most important value to consider is the real part of the input impedance. The power transferred to the secondary is Ipri^2 * real(Rinp). Since primary current is a limited resource, real(Rinp) should be as large as possible. It should not be too large, since input voltage is also limited. You might not be able to get enough current into the primary then.
The value of the real part is shown in my previous diagram as the blue curve. It is basically the secondary resonance curve, its width given by secondary Q and its position by the secondary res frequency. The height of the curve depends on the coupling, secondary Q and primary Z.

A secondary MMC is a pretty idea, but it has a drawback. It will decrease the dependence of secondary fres on the arc load, but it will also increase secondary Q, making the blue resonance curve narrower. So it will decrease the input impedance if you are operating away from the secondary fres and increase correspondingly primary current.
Upper pole operation with the primary tuned low has the nice effect, that the pole frequency will drop with the secondary fres as the arc grows. So the coil does not become more detuned at higher powers.

Good luck with your bigger MMC. I hope it will show the desired effect.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on May 30, 2020, 11:32:37 PM
Uspring, two points of clarification:

I am seeing some variance in current between pulses with different spark branching, but its only ~+-10%

Running through my collected data again, the toplaod clamping voltage seems to be closer to 70kV. I was modeling the clamping voltage with a voltage source and diodes which was causing some weird effects. Fitting a resistor value to the waveform I was seeing gives a value closer to 70kV


I ran the coil outside last night with a 13nf instead of the previous 10nf and a ~15ms burst length over the previous 10ms.

The MMC adjustment did lead to a decreased secondary current, but it was too much of a decrease and spark length pretty strongly suffered. My simulation shows the new tuning would reduce primary current by ~3.5dB with the loading values for the spark length I was seeing previously. I made some mistakes in my test setup for this run and don't have good data, but I was seeing a tank current reduction a bit more than that. Given that there is a feedback effect with tuning and spark length this makes sense. Results were a bit disappointing, but its nice to verify a correspondence between the model and reality!

I ran through my model splitting the transformer into parts using a voltage and current source and it has a 1:1 correspondence with the old model using the (traditionally used) coupled transformer. Aside from the input impedance, the values I was posting plots of are not things you can directly measure in the standard model without some additional math. All that the behavior voltage / current sources do is model an ideal transformer with infinite magnetizing inductance so I can independently measure the current circulating in the magnetizing inductance.

It makes sense that it is not worth worrying about the current circulating in the current flowing into the magnetizing impedance makes sense; if I am operating at resonance all the power needs to go somewhere. However, what I think is the real optimization is somewhat related: the real component of the secondary impedance as reflected to the primary side at resonance (impedance of everything to the right of what I have labeled as NODE_MATCH)  and the resistance of the primary coil + primary MMC. This determines the "energy transfer efficiency", which represents the energy. 



I think my assumption of 0.5 ohms for the primary side resistance may be a bit pessimistic, but the model for the 10nf tuning reports 3.5 ohms as the real component of the reflected secondary impedance and the model for the 13nf tuning reports 5.8 ohms, which gives an "energy transfer efficiency" of 3.5/(3.5+0.5) = 87.5% and 5.8/(.5+5.8 ) = 92.0% respectively. It might be worth the time to formalize this in terms of the primary and secondary Q and characteristic impedance, but with my current values these differences seem minor. I am trying to find an explanation for some claims made by Steve Ward on 4hv that he was able to achieve ~ 20% less tank current for the same spark length with different tuning on his QCW tesla coil, and this seems like the most promising explanation that does depend on weird spark behavior, only the behavior of the coupled LC system itself.


If I analyze this further it might provide some more justification for the secondary MMC (and a lower impedance secondary). Right now the main motivations for the secondary MMC are that it would limit spark de-tuning, and a bit of cargo cult tesla coil'ing. Most other QCW tesla coils have run with a larger topload (Steve Ward used a secondary MMC for his portable QCW coil) but I don't want a larger topload for my coil due to aesthetic reasons. Also, I can use the secondary MMC as a voltage divider to monitor topload voltage, which would be cool. Q will go up, but the extent of the de-tuning should go down.

After testing with the 13nf MMC I had everything set up and wanted sparks so I cut out the new 3nF section and ran the coil with the original 10nF mmc and the new 15ms burst length, but at the 525V bus voltage. It was pretty consistently hitting the OCD at ~10ms but it did give me some nice sparks, with a new length record. Definitely over 3' and longer than my previous coil! I think the next coil optimization to make would be to change the state machine for my switching waveforms to hard switch the bridge legs alternately. I am switching loss dominated so it would close to double my thermal margin. The changes in the verilog should not be that complicated but I will probably need to solder on all the test points for the logic probe again, which will be a bit annoying.

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on June 01, 2020, 05:27:08 PM
I guess I was a bit too critical about the simulation in your post from May 29th. The circuit topology looks good, but in my calculation the upper pole seemed to sit at about 330kHz compared to 290 kHz in yours. You probably used a larger than 9.9nF primary MMC value and a bit larger arc loading for your diagrams.

Quote
I think my assumption of 0.5 ohms for the primary side resistance may be a bit pessimistic, but the model for the 10nf tuning reports 3.5 ohms as the real component of the reflected secondary impedance and the model for the 13nf tuning reports 5.8 ohms, which gives an "energy transfer efficiency" of 3.5/(3.5+0.5) = 87.5% and 5.8/(.5+5.8 ) = 92.0% respectively. It might be worth the time to formalize this in terms of the primary and secondary Q and characteristic impedance, but with my current values these differences seem minor. I am trying to find an explanation for some claims made by Steve Ward on 4hv that he was able to achieve ~ 20% less tank current for the same spark length with different tuning on his QCW tesla coil, and this seems like the most promising explanation that does depend on weird spark behavior, only the behavior of the coupled LC system itself.

In your 450 V run you reported a 80A peak current. That translates to an input resistance of 4/pi * 450V / 80A = 7.2 ohm. The modelled value is 3.5 + 0.5 ohm. I've adjusted arc load from 140k to 70k. That increases the modelled value to 6.0 + 0.5 ohm.

I have a strong belief, that arc length (for a given frequency and ramp up speed) only depends on power and power is given by Ipri^2 * real(Rinp). So he must have had a larger Rinp to balance the effect of a smaller primary current. I've no idea how he could have increased Rinp without the corresponding loss of primary current such as you have seen, when you added some capacitance to your MMC.
There is optimization potential wrt the 0.5 Ohm primary side resistance, but it is limited, as you say. You could, e.g. keep the 13nF primary capacitance and reduce primary turn count, while keeping the coupling as large as possible. That will shorten the primary wire length and reduce its resistance. But I have some doubts, whether this is worthwhile.
You can also reduce secondary Q, i.e. more turns, less toroid. That will also increase real(Rinp) so that there is some leeway to reduce primary Z. As before you can maybe save one or 2 primary turns. Again that seems hardly worth the effort.

Quote
...Q will go up, but the extent of the de-tuning should go down.

Be aware, that arc capacitance is only one half of the issue of a perfectly tuned coil. The variation in arc resistive load impacts the power transfer to the secondary in a similar way.

Very nice sparks you got there ;D
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on June 05, 2020, 12:54:06 AM
In your 450 V run you reported a 80A peak current. That translates to an input resistance of 4/pi * 450V / 80A = 7.2 ohm. The modelled value is 3.5 + 0.5 ohm. I've adjusted arc load from 140k to 70k. That increases the modelled value to 6.0 + 0.5 ohm.

Thanks for catching that. The L / C components all seem to match, I think any discrepancies with that are due to me switching things around between the different simulations. I did mess up the calculation of my primary impedance and end up with 140K instead of 70K secondary load resistance though, I have fixed that in my recent simulations.

I have been looking to optimize my new secondary / primary design so I have been doing some sweeps of impedance over loading. I dont think this directly correlates with spark length, but I am sweeping spark resistance as 210k/X and spark capacitance as 2.3p*X where X is ~ spark length and normalized to my precious test data, assuming that impedance value corresponds to a spark length of ~3ft.

Here is a sweep of my the coil as modeled as it was set up for my may 19th run ( green is the first sweep with SPARK_LEN=1, blue is the next, etc):




From this sweep it is clear my coil is running on the upper pole and that the upper pole is dominant. I have not formalized the math yet, but based on this sweep and some other sweeps it appears that when you are running at the upper pole with the upper pole dominant (secondary tuned below the primary), increased streamer loading will lower the impedance. This will lead to the "wine shaped" primary current where primary current rapidly ramps and make it relatively difficult to properly match the system to the capabilities of the inverter. This is somewhat evident for my 450V current waveforms and was really obvious for my 525V testing, where the current ramped much faster than the bus voltage. 

When running lower pole dominant the upper pole impedance will increase with spark length. This configuration requires a PLL based controller or a startup oscillator, but it seems pretty promising for keeping a ~ constant inverter current .


With this in mind I have started designing my new coil. My inverter can go up to 525V and 80A, so this gives a target impedance of 8.35 Ohms (18.4dB on these plots) for full spark length.

Based on some running a number of simulations and seeing what secondary configurations are possible in FEMM I have arrived at a a 9" winding of 27 AWG wire on a 3" diameter fiberglass tube with a 11pF secondary MMC and my existing 8"x2" torroid.

Playing around with the simulation I have arrived at this tuning like this, which gives me close to 18.5dB, with impedance increasing as spark length increases. I suspect that ideally I would tune the coil so the impedance reaches its maximum right when maximum spark length is achieved. Overall this tuning seems pretty promising.





Tuning the system exactly for maximum spark length seems a bit challenging given inevitable (but hopefully minor) differences between simulation and reality. I think I am going to need a free parameter (or two) to shift the frequency and/or magnitude of inflection point of the impedance with spark loading. I tapped primary seems a bit ugly, but that would work. A small-ish (2uh or so) external inductor and spacers to change my coupling factor should also work. 

As a progress update, I ordered the secondary former today from a model rocket supply place, I am going to have to source some fiberglass epoxy next. I am going to have to do some more experimenting in FEMM to see what range of coupling factors I can achieve. I am hoping to still use my existing 10nF MMC, but I guess that could change too.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on June 06, 2020, 06:46:57 PM
That certainly looks like a sound design to me. I had a look at the primary current file you dropboxed. Thank you. Any chance that you could add input power information? The set of current amplitude, frequency and input power values might make it possible to extract arc loading capacitances and resistances. Particularly, since you have established accurately your tank parameters. I believe your estimate (210k/X and 2.3p*X) to be close to reality, but maybe it needs some fine tuning and it would be nice to establish the dependence of X on input power.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 12, 2020, 05:18:51 AM
Weston, first, excellent job with this build and development, good to see more SiC getting put to good use ;-).

I'm noticing that you model your spark as a parallel RC, but as far as i know, everyone else is assuming a series RC model, so any numbers based on my postings are always series RC.  The 210k/X and 2.3p*X seem pretty much in line with my estimates, but try series RC connection, i think its a better representation of the real thing.

I find i generally agree with Udo's take on the mechanics of the QCW DR system, i really enjoy reading his posts as they help me understand things better as well! In particular, on this subject of tuning:

Quote
Better is:

---flowerpole----fpri---------------fsec----fupperpole----      ; initially
---flowerpole----fpri-------fsec----fupperpole----      ; medium power
---flowerpole----fpri-fsec----fupperpole----      ; high power

It is a bit weird, but in upper pole operation, with the primary tuned low, it does not really matter at which frequency the primary tank resonates, as long as it is below the secondary fres.

My best guess for Qsec comes from Wards top voltage measurements with his sword sparc QCWs. Typical voltages were around 70 kV peak and only weakly dependent on length of the arc. You can derive a ballpark arc loading resistance by using your power level and this voltage value.

The way i like to think about this mode of operation is by comparison to a "SSTC" of the "single resonant" or "secondary resonant" type which looks like a series resonant secondary fed by an inverter via a series inductance (primary leakage).  We can improve the SSTC by canceling some of its primary inductance (the leakage part) with a capacitor, improving the power factor for the inverter.  You can think of this extra capacitor as only "partially" cancelling the leakage inductance when Fpri<Fsec, giving the system the extra Q boost when spark loading is very lossy.  If you let Fpri = Fsec, under heavy spark conditions, you have the optimal setup, i'd argue, for that operating point.  You can get into trouble, however, if the primary impedance is "too low" for your inverter, it will demand too much current as you get near Fpri=Fsec.

One small caveat to the claim above is that if your secondary Q drops too low, the reflected resistance on the primary side will limit the current, and the system may cease to even oscillate at the secondary resonance (aka "upper pole").

I also found some notes about spark voltages with my Teslagun V1, first some specs, then some records of voltage vs branchless spark length.

Cpri = 18.25nF
Lpri = 17.8uH
Csec = 27.3pF (including 16pF internal MMC)
Lsec = 8.23mH
K = 0.31
Breakout (initiate) Voltage: 39.4kV at Ipri = 48A @ 375khz

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 12, 2020, 05:18:09 PM
Quote
I have a strong belief, that arc length (for a given frequency and ramp up speed) only depends on power and power is given by Ipri^2 * real(Rinp). So he must have had a larger Rinp to balance the effect of a smaller primary current. I've no idea how he could have increased Rinp without the corresponding loss of primary current such as you have seen, when you added some capacitance to your MMC.

I think the apparent performance gain here was partly under-estimating the cost of branches in the sparks, and the tuning change promoting branchless sparks.  I'm assuming whatever claim of 20% less primary current for same spark length was from tuning the primary resonance below secondary and using the starting oscillator to force the mode to secondary resonance (upper pole).  This likely resulted in a straighter spark with less branches that made better use of the power, is my guess at this point.  There is some transient nature to these sparks, afterall, they behave "poorly" if not fed at an optimal rate.  Once the ramping has reached its saturation level, the spark might have nearly the same input power, but will transition from a long spike into a short serpentine with thicker "arc" channel.  My guess is that tuning alters the "impedance trajectory" that a coil operates at over the ramp duration, in ways that can be more or less beneficial for growing long branchless sparks. 
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on June 12, 2020, 07:23:47 PM
Hi Steve, it's certainly a pleasure to hear from you and draw on your expertise here.
Quote
I'm noticing that you model your spark as a parallel RC, but as far as i know, everyone else is assuming a series RC model, so any numbers based on my postings are always series RC.
I've used a parallel circuit, since it allows me to lump the arc capacitance into the top load capacitance. That simplifies the equations a bit. The series circuit is somewhat more intuitive as it reflects the arc being a resistive medium feeding a space charge. The difference between both types of circuit is small, though, if the frequency range over which the coil operates is limited.

Thank you for your measurements. The voltages seem rather low. Are that peak or RMS kV? The supply voltages you quoted seem more or less constant. Did you also use a phase shifted bridge? Otherwise, how do you control power input?

Quote
I think the apparent performance gain here was partly under-estimating the cost of branches in the sparks, and the tuning change promoting branchless sparks.

The upper pole operation with the primary tuned low shows a much flatter response of the input resistance when arc load increases. The other tuning often leads to a steep drop in input impedance at the end of the burst, so that power input suddenly rises. That could cause branchings.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 12, 2020, 08:57:35 PM
Quote
Thank you for your measurements. The voltages seem rather low. Are that peak or RMS kV? The supply voltages you quoted seem more or less constant. Did you also use a phase shifted bridge? Otherwise, how do you control power input?

I'm usually more of a "peak" value kinda guy. I'm assuming those are the peak voltages i recorded.  And yeah, they do seem low.  Maybe its worth taking another shot at calibrating the probe - i noted my last check was 758:1 where i think the nominal ratio is supposed to be ~1000:1.  However, i did record the secondary coil base current, and when i run my spice model, i see pretty close agreement to the measured primary current, secondary current, and top voltage.  I'll update the attachment in the previous post with base current measurements.

Yes, this setup used phase shifted bridge control.  The bus voltage reported is likely recorded at the very end of the ramp.

I do recall something like 70kV maximum for my first QCW project (because i recall pushing the 60kV rated probe to the point that it would occasionally emit a glow discharge internally) which did make larger and more branched sparks at full power relative to the measurements reported on the TeslaGun V1 system.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on June 12, 2020, 09:21:43 PM
Uspring: I dont have any more data from that rub beyond that I started with a DC bus voltage of 450V. I will have to start measuring the input voltage too. Its a bit of a pain to set up everything for testing because I have to do it in the open courtyard of my apartment complex.

Hi Steve! Thanks for dropping by.

I think I am pretty close to having the design for my new primary / secondary complete. USPS seems to have lost the package with my coilform though....

One QCW theory question:

Whats the deal with coupling? From theory and my simulations changing coupling alters the impedance of the system and the ratio of power circulating the in the primary and secondary. It seems that aside from practical / standoff voltage constraints a higher coupling would be desirable  Your first QCW coil had very high coupling but your tesla coil gun only has a coupling of 0.31. In some old 4hv posts you commented that the system seemed to perform better with a lower coupling. Dr. Killivolts coil uses ferrite to get K=0.55 https://highvoltageforum.net/index.php?topic=1073.0 .

Is there some optimization or some factor I am not considering or something? Once tank impedance is normalized is the coupling actually not that important as long as it is not excessively low? Given the high Q of the primary it seems that for most tunings the power dissipated in the primary + mmc will be minor compared to the power delivered to the spark.


I have been thinking about the QCW tuning and the d(Impedance)/d(Power) due to spark loading. When the secondary is tuned above the primary the impedance at the upper pole increases with spark loading, when the primary is tuned above the secondary the opposite happens. If impedance decreases with power you have a positive feedback effect that makes it difficult to control your rate of spark growth. I assume this can make it difficult to grow straight arcs.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 13, 2020, 07:26:46 PM
Coupling is sort of confusing to me, too.  I think about it as controlling the how much of the spark load we want reflected onto the primary, which should increase real(Rinp) relative to Zpri.  I think the answer is right there in Uspring's post on may 30? 

I think it comes down to your assumption that the component losses are negligible.  If that is indeed the case, then the payoff for higher couplings is less and less (i think this is probably right, but if its trivial to raise the coupling, its like free efficiency boost, even if its small).  In the case of my teslagun V1 i had flashover issues and also the spark growth was branchy and inconsistent with the higher K.  So i still think there is maybe something to this tuning stuff, which is only important with respect to spark growth/power control behavior, and not at all important in terms of power efficiency.
Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Uspring on June 13, 2020, 07:35:52 PM
Steve, thank you for the update with secondary base currents. A quick look at them seems to confirm, that your top load voltages indeed are peak voltages. I'm working at a more rigorous check. In the back of my mind I always have an improvement of my arc model. QCW data is particularly interesting as these arcs are easier to understand than the short lived usual DRSSTC bursts, since there the warming up of the arc channel has to be considered. Steady state QCW arcs avoid this problem.

Weston wrote:
Quote
Is there some optimization or some factor I am not considering or something? Once tank impedance is normalized is the coupling actually not that important as long as it is not excessively low?
A big coupling constant makes the system less sensitive to arc detuning. Consider the 2 tanks at the same resonance frequency i.e. fsec=fpri. The the poles will be at fsec/sqrt(1+k) and fsec/sqrt(1-k). A large k will move the poles (i.e. operating frequencies) away from the secondary resonance frequency and will actually detune the system. But this is compensated by the fact, that a large k will also increase the primary impedance. In many cases, these effects will more or less cancel each other in terms of input impedance. Look also at the equation here: https://highvoltageforum.net/index.php?topic=1024.msg7498#msg7498

But: Since the coil with the large k is already detuned considerably, additional detuning due to the presence of the arc doesn't matter as much anymore.
This is a bit of a ballpark estimation, since input impedance depends much on the location of primary and secondary resonance frequencies and the pole you're running at.

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 14, 2020, 07:52:40 PM
On the topic of switching losses, I'm interested to know what your gate driver solution looks like, if you've opted for negative bias and what the effective Rgon and Rgoff is.  One thing that i wonder about is why SiC datasheets seem to imply a minimum external gate resistance without specifying it.  What i mean is, all of the data graphs wrt gate resistance stop at some fairly high value, on the order similar to the internal Rg_int.  Seems to me there is still potential to switch off twice as fast as the datasheet gives data for.

On my 3 phase SiC Coil (which, I should make a post about) I'm using boot-strap gate drive at ~20V with no negative bias.  My gate drive IC can sink 10A even down to a few volts, and i use no external gate resistance on the C3M0021120K transistors, which are quite similar to the module you use.

Another idea to consider is a small amount of capacitive snubber across the switches, as this can greatly reduce switch-off losses at the cost of requiring a minimum current/deadtime for ZVS turn on.  If you check the spec sheet for Eoff vs Id, you can see the effect of the junction capacitance as a snubber- instead of a linear increase in switch off loss versus current, its more exponential since low current switch off is effectively snubbered by Coss.  So even a relatively minor boost in Coss can have a nice improvement in Eoff, so long as we dont have to pay for it with Eon (and with ZVS, that is zero).  In the scenario of using extra capacitor snubbers, it may make sense to keep all the hard switching to 1 bridge leg instead of alternating, so that the ZCS leg doesnt have any extra cap, which limits ZCS performance somewhat.

Since you estimate half of your losses are from switching, these might be nice options.

Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Weston on June 17, 2020, 01:56:22 AM
I am using UCC5390SC isolated gate drivers, which are a 10A minimum, 17A typical gate driver, with a 1 ohm turn on and 1 ohm turn off gate resistor. Its powered by a +15/-3V isolated converter. So it seems I should be getter pretty close to the datasheet switching speeds.

I dont think its quite likely with the expected switching speeds, but I am a bit worried about exceeding the CMTI of the isolated gate drivers, which could cause things to blow up in a hurry.

The negative gate bias helps prevent dv/dt induced turn on. In the module there is a bit higher gate inductance due to the packaging so the effect clamp impedance is higher. I don't think its much of a worry for these applications and newer devices are better in this respect, but the SiC MOSFET gate threshold can shift down with aging. I think on some early devices it could even make it to 0V.... So I suspect thats the main reason for the emphasis on negative gate bias. Either way, a +15/-3 isolated converter was only a bit more expensive than a +15V converter, so I figured why not.

I considered adding a capacitive snubber for turn off but it just seems to increase the potential for things to go wrong. Conduction loss goes as I^2, so even if I were to substantially reduce the switching loss I would only get a bit more current out. Alternating the switching leg is just some more verilog and would half the per FET switching loss without any hardware changes. Anther avenue would be to have some sort of adaptive over current protection based on the total switching losses. Right now my OCD is set assuming hard switching at 80A over the entire burst, which is not a realistic assumption. Davekni was discussing an analog implementation of this a while ago, I have  ~unlimited DSP power on the FPGA so it would just be a matter of coding it.

I have been a bit busy with the impending end of the spring term at school. I just ordered some PCBs to serve as FR-4 endcaps for my new secondary ( I am going to try the whole fiberglass secondary thing...) and my secondary MMC PCBs. Hopefully I will be making more progress soon-ish, term ends this week.


Steve: you should create a post on your 3 phase coil, I saw the youtube video a while ago and it seems pretty cool!

Also, do you have any guidance on the maximum per pulse junction temperature rise? I was reading some app notes related to it, but it seems most guidance is guided towards longer time period thermal cycles and infineon does not actually provide any of the required values to calculate thermal aging for the SiC parts. I arbitrarily chose 50C for my max delta T, mostly based on maximum junction temperature and some worst case assumptions on heatsink temperature. It seems like it might be high, but its also driving the calculations for my assumption of hard switching at maximum current for the entire pulse, which is also a bit unrealistic .


Title: Re: Portable Q(uarantine)CW Tesla Coil
Post by: Steve Ward on June 18, 2020, 02:14:42 AM
Sounds like a nice driver setup, no room for faster switching there!  One thing i might warn about is taking care with fast turn on, in the event the body diode does have reverse recovery, you might consider a bigger Rg-on to reduce the di/dt of such an event.

I ran into a few problems with my system, one of them somewhat related to CMTI (common mode transient immunity - how fast can you switch the high voltage side of a gate driver before causing it to malfunction, in possibly catastrophic ways), but really more of a input noise coupling issue that was solved by boosting my PWM input signal filter cap from 150pF to 1nF.  After resolving this (and other issues), I checked things out by switching 900V @ 175A, which produced about 100V/nS slew rate, double the "typical" rating on the silicon labs SI8231 that i use for isolation/deadtime generation.  I also checked for dv/dt induced turn on, looks ok, even with my 0V turn off.  However, dv/dt induced turn on shouldnt really be a thing (i mean, yes, it slows down the turn off) if you're switching ahead of the current zero crossing.  I also check the hard diode recovery - even though it shouldn't happen, it is important to be capable of handling it.  No noted problem with 900V 100A hard switching, but I do use 4.7 ohms for Rg-on to tame things a little.

I've had noise coupling issues on a number of SiC and GaN inverters i've developed, so to me this stressful switch testing seems crucial to weed out noise feedback problems that only happen when the voltage or current is high enough. If i can, i like to go about 20% over normal voltage and up to 2X the normal switching current.  Of course, if i blow up some parts it's not coming out of my pocket, so its a little easier to go forth with such tests.

Pulse die heating is probably a big driver of unreliability, but i dont have a good feel for what the MTBF versus delta T looks like for any of this stuff, and i think no device manufacturer will confidently post specs on this kinda thing.  When i used to work at Fermilab, which has lots of pulsers, the guidance was to limit the pulse delta T to just 3 degrees C!  However, those machines are typically pulsing at 10s of hz for ideally... 10..20...30 years and its really expensive to shut stuff down to fix it.  When i work out the delta T for the IGBT tesla drivers its not uncommon to see a worst case of 30-50*C, however, thats also under some of the assumptions you made, like hard switching the maximum current for the entire ramp, so really its maybe half of that delta T.

Aging of semiconductors just seems weird, and beyond my 1 semester of semiconductor physics understanding. I was completely unaware of this Vgs-threshold change with age.  I did run into a reference on SiC body diode Vf increase with age, but as far as i can tell that's probably not an issue with recent SiC devices.

SimplePortal 2.3.6 © 2008-2014, SimplePortal