Eliminating Shoot-Through issues by Adding Dead-Time

Got friendly advice? Post it here!

Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Thu Feb 05, 2015 11:22 pm

On my oneTesla coil, I have been fighting some issues of "shoot-thru" in the half-bridge driver that causes both transistors to conduct and short, blowing both transistors and the fuse and sometimes frying the resistor on the gates and other carnage.

So, I designed this little add-on this past weekend. It has greatly helped by allowing for an adjustable dead-time between the pulses to the half-bridge. The circuit uses two chips and fits in series between pin 4 of the 74HCT14 and pin 2 of each of the UCC2732x driver chips. You do have to swap-out the UCC27321 for a second UCC27322 as this circuit needs two non-inverting drivers instead of one inverting and one non-inverting.

I'm posting it here for others who have had similar issues. It doesn't fully solve all failure modes possible, but definitely helps. The schematic is attached -- sorry for my poor handwriting and sloppy sketching (I should redraw it), but this is simply a page I scanned out of my lab notebook where I doodle up all of my designs and keep my notes organized.

PWMMainDriver.png
PWM Add-On for oneTesla Main Driver
(408.89 KiB) Not downloaded yet


Tools needed: Oscilloscope with at least two channels and delta-T markers, signal generator capable of producing either a +5V TTL signal or a +/- 5V (or so) AC sinewave in the 200-300kHz range, and a frequency counter (could use the delta-T markers on the scope).

To use and calibrate:

1) First, use the original circuit as-is with a scope on pin 4 of the 74HCT14 to find the frequency of the primary circuit with the coil running generating high-voltage (Warning: Care must be exercised to isolate test equipment and keep output power minimal!!). This has to be done using the delta-T markers on the scope instead of a frequency counter because the interrupter will not have the coil running continously (and you wouldn't want to be that close to it measuring frequencies if it was). You have to trigger the scope at the right spot and crank up the intensity of the beam to handle the rescans necessary to see it (i.e. it will be dim) on analog scopes. On digital scopes, you may be able to program it to simply record the correct spot of the wave (I'm old school and tend to prefer using analog scopes, even though I own both). For reference, mine measured 286 kHz.

2) Build and Connect this add-on circuit between pin 4 of the 74HCT14 and pin 2 of each of the UCC2732x driver chips, replacing the xx321 driver with a xx322 driver. This new circuit needs two NON-inverting drivers to function. The original oneTesla design has one inverting and one non-inverting. So you have to use two UCC27322 chips instead. You'll need to either cut traces or pull the pins out of the sockets and solder a short lead to connect them as the original circuit tied both pin 2s of the drivers together. This circuit needs them separated. And you must disconnect the existing route of pin 4 of the 74HCT14 that's currently going to pin 2 on both drivers, as that pin must go to this circuit first and not to the drivers. Note: The output of pin 4 of the 74HCT14 must continue to connect to the clock input of the 74HCT74 so you do not want to pull its pin out of the socket. Instead, solder on a wire either at the feed-through pad that's on one of the trace-routes on the board or simply solder it to the pin with the pin still inserted in its socket, which is what I did.

3) Power-up ONLY the low-voltage circuitry and put a dual-trace scope on pins 11 and 14 of the SG3525 (or pin 2 of each of the drivers). Adjust the dead-time potentiometer (pin 7 of the SG3525) to set the desired pulse separation for the drivers, but don't go too width with the separation or you'll run out of bandwidth on the SG3525, as it's only rated for 400kHz for the oscillator (which must run at twice the resonant frequency of the primary winding). Note: If you set it too wide, you'll find out in the next step as you won't be able to set the desired frequency without the SG3525 outputs going "Minnie-Mouse" on you. If you later want to change the dead-time, you must also readjust your frequency in the next step as the dead-time pot affects it too. Even at the minimum setting, you'll get a nice dead-time added. The exact range of dead-time you can get depends on the exact frequency you are running. For reference, my circuit (running at 286 kHz) is adjustable from 266nS to 792nS as indicated in my chicken scratch notes that I determined while calibrating mine. I'm actually still running mine at around 266nS, which is significantly more than the original oneTesla circuit, which only had the propagation delay of the drivers to add dead-time.

4) Adjust the frequency set potentiometer (pin 6 of the SG3525) until the frequency is about 5% lower than the target resonant frequency of the primary. For example, I tuned mine to 270 kHz for a resonant of 286 kHz (meaning the output pins 11 and 14 of the SG3525 should indicate 286 kHz on the frequency counter and pin 4 of the SG3525 should read about 540 kHz, yes it exceeds the 400 kHz published limit of the SG3525, but it still works well. My chip here actually functioned correctly all the way to the mid-600 kHz range, though the higher it is, the less adjustment you'll have on the dead-time control).

5) Hook a signal generator to the current sense transformer conditioning circuit on either side of the 0.1uF cap (depending on the type of signal you are generating -- one side needs a sine-wave, the other can use a square wave)...

6) Set the frequency on the signal generator for the resonant frequency measured in Step #1. Verify a corresponding signal on pin 4 of the 74HCT14. It should be a nice square-wave at your resonant frequency.

7) Use a dual-trace scope to display one of the two outputs of the SG3525 (either pin 14 or 11) and pin 3 of the SG3525 (or pin 13 of the 74LS123), which is the sync signal and will be the 2x clock source to do a phase-locked-loop on the SG3525 so that it creates the same frequency as the resonant of the primary. You should now see the SG3525 output pins resyncing in conjunction with the pulse coming into pin 3, causing the frequency on pins 14 and 11 of the SG3525 to shift accordingly, forming a simple phase-locked-loop circuit.

8) Adjust the sync-pulse width potentiometer (pin 15 of the 74LS123) until the SG3525 is able to correctly lock onto the same frequency as the signal generator is supplying. You should find that there's a band in the adjustment range of the sync pulse-width where it will pull-in to the correct frequency and lock to it. Adjust it to the middle of this adjustment band so that there's ample tolerance on either side in case the frequency difts slightly.

9) That's it... Disconnect the test equipment, and use the coil as before...

With this circuit, I've been blowing far fewer transistors than before. If calibrated correctly, output power should be the same, if not better than without it. And if you needed to remove it for some reason, you can simply disconnect this board and jumper the pin 2 of the drivers back to their original connection point and replace the non-inverting driver with the original inverting version.

Also, I've found that 3-position Phoenix Contact connectors (normally used for attaching wires to a circuit board) will fit the transistor holes and allow a method to "socket" the transistors. After the 5th or 6th time of unsoldering and resoldering the transistors (a delicate, time-consuming task), I switched to that and now I can change the transistors without even removing the circuit board. Though with this circuit addition, I haven't needed to do so quite as often. :-)

New parts required:
- second UCC27322 to replace the UCC27321
- SG3525 IC
- 74LS123 IC
- 2x 16-pin sockets
- 2x 1N4148 diode
- 2x 1K Resistor 1/4W
- 2x 470-ohm Resistor 1/4W
- 1x 100-ohm multiturn potentiometer
- 1x 2k-ohm multiturn potentiometer
- 1x 20k-ohm multiturn potentiometer
- 2x 220pF cap ceramic
- 1x 22pF cap ceramic
- 2x 0.1uF cap ceramic or mylar
- 1x 0.001uF cap mica
- 3x 0.01uF cap ceramic (decoupling, not shown on schematic, tie two on the +5V supply near each IC and one on the +15V supply near the SG3525)
- 2x 10uF cap elect. (power filter, not shown on schematic, tie one across the +5V supply and one across the +15V supply)

(Note: it may work OK with a 74HCT123, but the specs on it are a bit different and I haven't tried it. I've only used the LS family part, since I happened to have one on-hand already)... So you'll have to experiment. Maybe someone can test that and post their results...

Also, you may can get away with a mylar cap for the 0.001uF timing capacitor. But it will have worse frequency drift than a mica part. As long as it stays within the 5-10% window of tolerance for the SG3525 sync logic, it should work. The mica cap is definitely better and I happened to have one on-hand, so that's what I used.

And, tie the unused logic input pins of the 74LS123 to either +5 or gnd (also not shown on the schematic).

Enjoy!
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Sat Feb 07, 2015 4:55 pm

A friend looking over my original posting was a little confused about the need for two non-inverting driver chips with this circuit. So I'm posting my reply to him here and including a more detailed explanation of the circuit function in general.

The requirement of two non-inverting drivers shouldn't be a mystery. The oneTesla Coil uses a half-bridge to drive the primary winding. The half-bridge uses two transistors to alternately drive one side of the coil to V+ and ground (one transistor takes it to V+ and the other to ground), meaning that only one of the two transistors can be turned on at a time.

In the original design, they used the square-wave output from the current sense circuit to go to the input of both driver chips and they had one inverting driver and one non-inverting driver. The non-inverting turned one transistor on when the square-wave was high and the inverting turned the other transistor on when the square-wave was low. This allowed it to toggle between the two transistors turning only one on at a time.

The problem is that transistors turn on quicker than they turn off. This means that there's a slight overlap between the on and off cycles if there isn't a slight delay between switching which transistor is turned on. If both transistors turn on simultaneously, then you have V+, which in this case is the rectified direct AC mains voltage, shorted directly to ground, a problem known as "shoot-through". That causes the transistors to fail.

Short bursts you can get away with, but it produces a lot of heat (and reduces operating efficiency and overall power in the coil) and if the overlap gets slightly longer, which can happen because they get even slower as they get hotter, then eventually --- boom! The transistors short out completely and die...

The only delay between the pulses in the original circuit design was the propagation time through the driver chips and gate drive transformer, plus the capacitance of the gates of the transistors (i.e. their charge/discharge time). Granted, the IGBT's used in the oneTesla driver design has pretty fast on and off times and nearly symmetric on/off times, which is what mostly kept them out of trouble. But still, if you load up the Tesla Coil a bit to where there is more current draw through the transistors, the switch-on vs. switch-off time differential would worsen, and they would eventually fail... And that's why I made this circuit.

The SG3525 requires a clock that's twice the frequency of the outputs its driving, so that it can use one cycle to turn on one output (one transistor) and the other cycle to turn on the other output (other transistor), completely independently. So, I used the 74LS123 circuit as a frequency doubler to double the square wave from the current sense transformer -- giving me a short pulse at the beginning of both the high-period and low-period of the original circuit signal. That gets feed to the SG3525's sync input to clock it and phase lock it to the current sense transformer (and lock it to the resonant frequency of the primary winding in the process).

The SG3525 has two outputs that can either go to a driver or straight to the transistors (depending on your circuit and needs). But the outputs are both high-going pulses for on. That means that you can't invert one of them before sending it to the transistor or you'll end up turning on both transistors at the same time (very bad). So you have to use two non-inverting drivers. That way each transistor is turned on only when it should.

And they have to be non-inverting because the SG3525 is setup to drive the outputs high for on and low for off. There's a SG3527 with them inverted. And on it, you'd need both to be inverting drivers to invert them back to the correct polarity before going to the transistors. Essentially the SG3525 is designed for driving N-channel devices (as these transistors are) and the SG3527 is designed for driving P-channel devices.

The SG3525 has a circuit that lets you use a resistor or potentiometer to set a delay time between the two output pulses, known as the dead-time. During this time, both outputs are off, allowing time for the one transistor to turn off completely before turning on the other. That was my whole reason for adding this circuit as it gave me a way of dialing in a precise nanosecond delay between switching transistors.

In my opinion, this "add-on" isn't really optional. They are just barely getting away with things in the original circuit design and if you need to switch to a different transistor, such as one able to switch higher voltage or current (as I have also been experimenting with), well, they tend to have even longer turn-off times (with a nearly same fast turn on time) which only exacerbates the problem even more.

I originally experimented with other tricks of adding dead-time. But they were all sub par in one way or another. For example, I tried things like a larger gate resistor with a diode so that the on-time gate charge time would be longer than the off-time gate discharge time to bring the on and off times back to being more symmetrical with one another. But that caused problems with the startup or "ring up" operation of the primary and tended to cause it to not want to start oscillating.

There were some other failed attempts to do similar tricks with the pulses in the current sense transformer buffer circuit. While I got several configurations to function, they were still too picky about the specific component values and exact operating conditions of the coil. So I finally punted and designed a proper circuit for it.

Hope this helps explain things...
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Fri Feb 13, 2015 3:07 am

Finally, I'm getting around to creating a much better schematic for this circuit. Something more legible than my chicken scratch hand-drawn version.

PWMDeadTimeDriver_eeschema.png
Better PWM Dead-Time Driver Add-On Schematic
(505.55 KiB) Not downloaded yet


Bill of Materials:

C1 220pF Ceramic
C2 220pF Ceramic
C3 22pF Ceramic
C4 0.001uF Mica
C5 0.1uF Mylar
C6 0.1uF Mylar
C7 0.01uF Ceramic
C8 10uF 25V Electrolytic
C9 10uF 25V Electrolytic
C10 0.01uF Ceramic
C11 0.01uF Ceramic
D1 1N4148
D2 1N4148
R1 470 Ohm
R2 470 Ohm
R3 20K Ohm Multiturn Potentiometer
R4 1K Ohm
R5 2K Ohm Multiturn Potentiometer
R6 1K Ohm
R7 100 Ohm Multiturn Potentiometer
U1 74LS123
U2 SG3525

Enjoy,
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Wed Mar 04, 2015 12:55 pm

Sub-Subject: Fine-tuning values to match your coil...

The values specified in the original Dead-Time Add-on Circuit worked well for my original coil configuration. But after I made some modifications, like adding a strike-ring and a special ground-plane, I found that the resonant frequency of my secondary had dropped nearly 7% from what it was, from 246.5.kHz with a simulated streamer (284kHz or so without the simulated streamer) down to 230kHz (260kHz without the simulated stream).

In my original "out-of-the-box" configuration, my primary resonance (which was found to be 286kHz) was really close to my secondary without the simulated streamer, but was 16% higher than the secondary with the simulated streamer. Now, typically you'd tune them intentionally out-of-tune slightly, but generally the other way around -- making the primary a little lower than the secondary so that when streamers form (especially ones that strike a ground) and pulls down the resonant frequency of the secondary, it moves more in-tune with the primary than out.

Nonetheless, my original experimentation was with the out-of-the-box setup with the coil built as-per the instructions. And so my original Dead-Time circuit was optimized for a 286kHz primary frequency. But now that I've made some physical changes and caused them to be a bit out-of-tune in general, I'm looking at modifying the tank capacitor and/or winding of the primary to correct my tuning issues.

What I'm finding is that the original component selection on my Dead-Time Add-On circuit is inadequate for the potentiometer R5 (2K) and fixed-resistor R6 (1K) values (see the better schematic picture in my 3rd posting on this topic) in selecting the proper frequency.

One solution is to change the potentiometer R5 from 2K to 5K. That should give you all of the adjustment range you'll ever need. However, you will sacrifice accuracy of fine-tuning adjustment. Therefore, you may opt to change the fixed-resistor R6 instead, which is my choice.

Here's a little chart showing common fixed-resistor values for R6 with the original 2K adjustment potentiometer R5:

|Zero Dead|Max Dead|Working Range|Effective Range
Rfixed|FreqMinFreqMax|FreqMinFreqMax|FreqMinFreqMax|FreqMinFreqMax
2700|151976264550|139276228311|151976228311|151976205479
2400|162338297619|147929252525|162338252525|162338227273
2200|170068324675|154321271739|170068271739|170068244565
2000|178571357143|161290294118|178571294118|178571264706
1800|187970396825|168919320513|187970320513|187970288462
1500|204082476190|181818370370|204082370370|204082333333
1200|223214595238|196850438596|223214438596|223214394737
1000|238095714286|208333500000|238095500000|238095450000


I've shown the frequency range for both minimum and maximum dead-time adjustment. The Working Range column shows the minimum and maximum frequency for the full-span of the dead-time adjustment (that is the maximum minimum-frequency and the minimum maximum-frequency). And the last column, labeled Effective Range, is the same as the Working Range, but has the maximum frequency scaled down by 10%, as I've found in nearly every situation that the real frequency is about 10% less than that calculated by the SG3525 timing equations. So that's the column you should use.

I'm sure this 10% degradation of frequency is a composite of the tolerance in the involved components as well as the SG3525 equation being only an approximation. The minimum would similarly be scaled down by 10%, but I left it unchanged as only the tighter range should be considered the real working range. Though, I've never seen it be off in the other direction with higher frequencies than those predicted, only lower frequencies.

On my circuit, I am changing R6 from 1K to something closer to 2K to allow me to target the new resonant frequency of my primary and to still allow for a nice range of adjustment without giving up fine-tuning ability on the potentiometer.

I also highly recommend the JavaTC applet found on http://www.classictesla.com/. I have found it to not only be extremely useful, but quite accurate as well, assuming you are careful in measuring things and entering correct dimensions and values. In fact, when I entered the data for my coil, its predicted resonant frequency for both the primary and secondary was only off by a few kilohertz.

The only value I question on JavaTC is the DC Resistance of the secondary. On mine, I measure 250.7-ohms. But it predicted 511-ohms, which is nearly a perfect factor of 2, and makes me wonder if they are simply missing a divide by two in their model. But otherwise, I found everything to be nearly exact -- that is, once I entered all of the new modifications I made to my coil, such as the strike-ring and ground plane positions and dimensions.

I hope you find this helpful.
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Sat Mar 07, 2015 4:16 am

Tonight while working on the tuning of my coil, I opted to disconnect and remove my Dead-Time Circuit and just get it out of the way temporarily so I wouldn't have to worry about recalibrating it while tuning the coil. This is easy to do on my setup because I put little header connectors on it for that very reason.

After several rounds of working with the coil tuning, I opted to play a song on it via MIDI and decided to give it a go without the Dead-Time Circuit installed. A little over a minute into the song and there's the infamous flash of green light of the fuse as the transistors die. I was playing "Amazing Grace", but am thinking I should have played "Another One Bites the Dust" instead, as my pile of dead transistors continues to grow.

I then replaced the transistors and fuse and swapped my Dead-Time Circuit back in and recalibrated it for the frequency of the coil. And then immediately played "Amazing Grace" again without changing any other settings.

I've made the videos available for you to see so you can compare them and see how the coil behaves with and without the Dead-Time Circuit installed. As I mentioned in a previous posting, if the circuit is properly calibrated, you'll not notice any performance differences other than your transistors won't die when running long pulse widths and higher powers. So I think this is a good demo of why I designed this circuit in the first place. It's amazing how much it helps just removing that little bit of overlap with a little Dead-Time.

Here's the links to the videos. The filenames will show which-is-which:

http://cloud.dewtronics.com/videos/oneT ... t_360p.mp4
http://cloud.dewtronics.com/videos/oneT ... t_360p.mp4

Happy Coiling!
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby Alex » Sat Mar 07, 2015 2:03 pm

interesting work!

I notice that you might be running more pulse width than is needed to get that spark length and that your coil is still slightly out of tune. you should be getting 20+ inches happily. Also, the MIDI file you played has a lot of high notes that really stresses the coil.

inspect the back of the dead igbt's, I've noticed in one of my failures that the back looked like they had almost melted??

do you have waveforms of the inverter output and current with your dead time circuit installed?
Alex
Incredible IGBT
 
Posts: 627
Joined: Sun Jan 13, 2013 12:16 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Sat Mar 07, 2015 3:42 pm

Alex wrote:interesting work!

I notice that you might be running more pulse width than is needed to get that spark length and that your coil is still slightly out of tune. you should be getting 20+ inches happily. Also, the MIDI file you played has a lot of high notes that really stresses the coil.

inspect the back of the dead igbt's, I've noticed in one of my failures that the back looked like they had almost melted??

do you have waveforms of the inverter output and current with your dead time circuit installed?


Oh yes, I'm well aware it's still out of tune. If I happened to have a 0.083uF cap on hand, it would be in tune. But one thing of importance I learned this past week is that it absolutely doesn't work to simply parallel caps together to form the primary tank cap. Doing so caused what looked on the scope as if 3 separate frequencies were being generated -- one for each capacitor and one for the composite of the two, as I clearly saw both 287kHz and 230kHz waves, along with one a bit higher that I couldn't get a read on, on the current switching logic with them paralleled. And the coil performed with nearly no output whatsoever -- just a little bit of plasma and corona discharge but no streamers (it killed it).

Once I find the right cap, which by my calculations based on measurements should be 0.083uF, I should be able to get it completely tuned. I'm opting for increasing the capacitor and dealing with the extra current generated by doing that rather than dropping the turns on the primary. I make that decision based on the results from JavaTC modeling of it, as well as perceptions of how things are currently working.

As for pulse width, a lot of that comes from the fact that I can presently run 8 simultaneous notes (and with a little doing run 16 simultaneous notes), as I'm not using the oneTesla interrupter, but one of my own design that uses four parallel Arduino ProMini boards.

And yes, that MIDI file has a bit of highs, but the coil should easily be able to span the full frequency spectrum supported on MIDI. If not, then I don't consider my work on the coil complete. I shouldn't have to alter what is being played to keep from destroying the coil.

I'm also working on designing a direct audio interrupter that can drive it directly from a mp3 player or smartphone. It's working, but I'm working on improving the sound quality by altering my modulation techniques. And now that circuit really taxes the coil quite heavily, even to the point of a continuous plasma spew instead of long streamers. Without the dead-time circuit, the coil doesn't have a chance with such a signal and dies after a few seconds. With the dead-time circuit in place, I can drive it that hard for many minutes and still not run into issues.

Nope, the backs of my IGBTs don't look melted. Visually inspecting them, they look fine. But with the meter, it shows a zero-volt drop between all three pins. Now, I have had them burn up one of the gate resistors in the process, but that has only happened once.

One of the things I was testing was the snubber capacitor that I had just added. That was a huge improvement to the overall sound quality of my coil and even increased the streamer length a bit and has definitely up'd the output voltage.

My next area of work on the coil, while looking for the proper size cap to tune it, will be to put a couple more coats of polyurethane on it and to space the secondary up from the primary a bit to reduce the coupling between them. The coupling coefficient is currently over 0.3 and I think should be more in the 0.1 to 0.12 range.

In a second MIDI test I was running last night on a much more dynamic but less "intense" piece (it was the 2nd Movement of Tchaikovsky's 5th Symphony -- one of my favorites), the increased voltage from the added snubber cap was causing some arc-through of the secondary to the strike-ring that I added sometime back. As soon as I saw that, I shutdown the test to keep from damaging the secondary winding. I think spacing it a bit to reduce primary to secondary coupling will help with that. And if not, it might require some geometric changes be made to the strike-ring. My design of the strike-ring was pretty much "seat of the pants", since I couldn't find any design guides anywhere on calculating optimal sizing, positioning, spacing, etc.

No I don't have waveforms captured of the inverter output or current with the dead-time circuit. Though on the scope, at least to me, they didn't really look any different than without it -- only with a slight bit more spacing between pulses (in the nanoseconds) and a slightly broader current envelope.

Overall, I have been pleased with the functionality of the dead-time circuit. In fact, I've found in my recent calibrating of it, that you can actually dial in a whole range of frequencies that it will phase lock to rather than targeting a specific frequency as I described in my previous posts, even though it technically only has to ever lock into the resonant frequency of the primary itself.

To do that, hook the SG3525 output to a scope and frequency counter and turn on the low-voltage/logic power (no high voltage). First set your desired dead-time width, as that adjustment affects the overall frequency and signal locking characteristics. I'm still running on the lowest setting on dead-time on this coil. Next, you set the desired lowest frequency you want it to be able to lock to using the frequency set pot, as can be noted on your frequency counter. This value should probably be based on the resonant frequency of the coil with a simulated streamer, as that feels like the best value to use.

Next, turn down the sync pulse width (increased resistance on that pot if my memory is correct) and apply a frequency with a frequency generator to the current sense transformer circuit, start with the same frequency you dialed into the frequency set pot. Then gradually crank up the frequency on the signal generator to the desired highest frequency of the range you wish it to lock, turning up the sync pulse width pot as you increase the frequency to the point that the SG3525 locks to that new frequency. Or to do it quicker, you could just crank it up to the highest frequency you want to lock and turn the sync pulse width pot until it locks.

At that point, you should be able to sweep the frequency on the signal generator anywhere between the low frequency set point and the higher frequency set point and the SG3525 should lock to it, only losing lock when you exceed that range on either side. I'm not sure the SG3525 was designed to operate as a phase locked loop, but it performs admirably as one for these frequencies.

I don't know how big of a range it's able to lock, but in that video, mine was calibrated to lock the range of about 250kHz to 300kHz. The primary, still with the original 0.068uF cap, is running around 287kHz as its resonant. The resonant of my secondary, without simulated streamers is 260kHz. With a simulated streamer it's around 225-230kHz. Once I get my hands on a 0.083uF cap, I believe it will be nearly perfectly tuned and I should be able to change my lock range to something more like 230kHz to 280kHz.

I'm soon going to be assembling another coil identical to this one so I can play in stereo. And to save having to hand-wire another dead-time board, I'm seriously considering drawing up a PCB layout and doing a board turn. I'm that satisfied with how well the circuit is functioning. I don't know if there is enough interest on here or not of others wanting to build the dead-time circuit, but if there is, I would consider a larger run of boards to sell to those that want them.

Regards,
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby Alex » Sat Mar 07, 2015 4:32 pm

surprising your circuit and coil can handle all of those simultaneous notes!

Why are you running your primary frequency higher than the secondary? Usually we let the streamer drop the resonant frequency down to the primary's frequency.
Alex
Incredible IGBT
 
Posts: 627
Joined: Sun Jan 13, 2013 12:16 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby dewhisna » Sat Mar 07, 2015 7:09 pm

Alex wrote:surprising your circuit and coil can handle all of those simultaneous notes!

Why are you running your primary frequency higher than the secondary? Usually we let the streamer drop the resonant frequency down to the primary's frequency.


Oh I don't want to run the primary higher than the secondary. That's how things worked out out-of-the-box from assembling it. The primary is 287kHz. And the secondary is 260kHz. Now, before I added the strike-ring and an extended ground plane, the secondary was closer to the 287kHz, not exact, but close.

I have similar Cornell-Dubilier capacitors of 0.068uF, 0.033uF, 0.022uF, and 0.015uF of the same 3000V rating and I was hoping I could simply parallel those on to correct the frequency of the primary. After much experimentation and calculations, I've concluded the ideal is to add 0.015uF, bringing it to 0.083uF total.

HOWEVER, what I discovered this past week completely surprised me. When I added a cap to it (and it didn't matter which one or what size), it absolutely killed the output of the coil if there was more than one cap there. I mean no streamers whatsoever, except maybe little tiny 3" to 6" things and that only if you cranked the pulse width settings to max. There was plenty of corona and a couple "spits" of plasma. It could light a fluorescent bulb, but nothing for correct operation.

I hooked up my equipment to measure the current sense transformer response (at the HCT14 output) with it running to see what the resonant frequency was and try to make sense of what was going on when there was multiple caps in parallel. And much to my surprise there wasn't a single set frequency there. No, it was bouncing back and forth between several frequencies.

When I had the 0.068uF and 0.015uF in parallel, for example, there was clearly a 287kHz of the 0.068uF cap itself. Plus, there was a completely separate square wave popping up around 230kHz that I surmise was the summation of the two (though it was less prevalent) and there was yet another frequency, a higher frequency that I couldn't get a good read on, that I think was that of the 0.015uF cap by itself.

In other words, with them in parallel, it wasn't behaving as a single cap, but instead alternated between them in the most unusual way. It was a total surprise to me and quite puzzling.

There's things that can be done to balance out the charge between the two to make them behave more as a single unit, but none that I know of that wouldn't adversely affect performance by introducing resistance into the equation.

So... since I have no 0.083uF capacitor on-hand, it's pretty much going to be running out-of-tune slightly until I can find one.

I've crunched through a lot of math and models like that of JavaTC and am opting for changing the capacitor rather than altering the primary winding. I know I could add between 1/2 and 1 turn of primary winding and it should bring it into tune, but I don't want to go that route.

So yes, I understand the benefits of running the primary at a lower frequency so that when the frequency of the secondary drops they come more into tune instead of going more out of tune... That's why I did the measurements of the secondary with a simulated streamer to determine approximately what the frequency should drop to.

I'm just waiting to get my hands on the correct capacitor...

And yep, it can handle all of those notes. I've played complete symphony pieces through it with all of the parts. Now, the harmony tends to drown out the melody when you do that, but yep, it can run full-bore for quite some time. Well, it can do it as long as I have the dead-time circuit installed. It wouldn't have a prayer to do that without it -- as you can see between those recordings of "Amazing Grace" that I posted of one with the dead-time circuit and one without.

Regards,
Donna
dewhisna
Tipsy Toggle Switch
 
Posts: 7
Joined: Thu Feb 05, 2015 9:48 pm

Re: Eliminating Shoot-Through issues by Adding Dead-Time

Postby Alex » Sat Mar 07, 2015 8:21 pm

Weird how your secondary came like that.

I recommend not adding more capacitance to the primary, as it would cause it to run higher currents. the secondaries are supposed to be around 320kHz, and with a ~12-15% streamer loading it should drop down to your primary frequency.

since the inductance of your secondary is rather large you could unwind some turns or wind a new one.

also if your coil is not experiencing flashovers, leave the coupling! :)
Alex
Incredible IGBT
 
Posts: 627
Joined: Sun Jan 13, 2013 12:16 pm

Next

Return to Tips and Tricks

Who is online

Users browsing this forum: No registered users and 0 guests


cron