Post by: profdc9 on May 08, 2018, 04:20:55 AM
I am working on a new version of the Universal Driver called UD2.9, which includes the pulse skip ability. By closing a jumper, pulse skip functionality is active, otherwise the standard overcurrent detection is active. I also added an external disable feature for adding, for example, thermal shutoff.
I have assembled and bench tested it, so far so good. :) Here's a picture of the assembled board:
[ Invalid Attachment ]
Hope to put this in the coil soon and not have blown IGBTs!
Dan
Post by: Mads Barnkob on May 08, 2018, 08:20:28 AM
A few questions that might lead to small optimizations:
- Could you move the electrolytic capacitors further away from the heat sinked voltage regulators?
- Could you space plastic parts further away from the large power resistors?
Looking forward to see it run a coil :)
Post by: profdc9 on May 08, 2018, 10:18:34 PM
I think I can probably move the burden resistors around though and not have to rearrange too much. I tried to design it for QCW like use, so I am aware that heat could be a problem.
Hopefully I'll get a chance to test it soon.
Dan
Post by: Hydron on May 09, 2018, 08:39:07 AM
I look forward to hearing the results of further testing - looks good so far (and sorry I wasn't able to be of more help scrutinising the design!)
Post by: profdc9 on May 09, 2018, 05:45:26 PM
Dan
Post by: profdc9 on May 12, 2018, 04:13:28 PM
So this morning I put the UD2.9 skip pulse in. I have not tested it yet with the secondary in, only the primary with some soup cans as a load. I am being cautious, so I set the overcurrent low (200 A) and the bus voltage low (30-40 V that is doubled).
Here is a trace of the voltage and current on the bus with the skip pulse jumper removed, so it behaves like an ordinary driver that trips and ends the pulse at a single overcurrent event:
Here I replace the jumper, and it allows the pulse to continue until the interrupter is over:
So far so good. The current is a bit rougher than I would like it but perhaps that is because I need to reduce the capacitance on the overcurrent detector comparator output, or just need to use it with a load. Hopefully I can try the coil tonight when it gets dark out.
Dan
Post by: Hydron on May 12, 2018, 04:42:12 PM
The current waveform isn't any rougher than what I'd expect from a DRSSTC primary tank with no load. You're right that it might look nicer once you have a streamer/ground strike as a load (system Q will be lower with a load) but in any case isn't something to worry about.
Post by: profdc9 on May 12, 2018, 04:52:45 PM
This circuit is the same one I sent to you except I changed a couple of component values. I blew two of the IGBTs with the old driver demonstrating it for some guests. It's getting hotter here (32-33 C) and I wonder if that's affecting how hard I can run the coil before the transistors fail.
Dan
Post by: profdc9 on May 12, 2018, 05:28:46 PM
This is without skip pulse, the jumper removed:
[ Invalid Attachment ]
This is with skip pulse, with the jumper replaced:
[ Invalid Attachment ]
Much better looking now.
Dan
Post by: profdc9 on May 13, 2018, 03:30:05 AM
/>
This used a Digispark programmed to be a OneTesla MIDI interrupter.
I think the main benefit of the skip pulse is that it allows the peak currents to be lower but still provide good arcs. For example, the bus voltage can be higher, and if the rise of the current is too rapid, the pulse is very short. The skip pulse mode lengthens the total time of the pulse.
As you can see, I do not have single streamers, but it did help me get some more power out of the coil I think. I'll release the UD2.9 design soon with the update to the DRSSTC pack.
Dan
Post by: oneKone on May 14, 2018, 05:45:45 AM
It honestly amazes me how fast your creating these boards.
Post by: Teravolt on May 14, 2018, 06:12:32 PM
Post by: futurist on May 14, 2018, 08:19:32 PM
Post by: profdc9 on May 15, 2018, 03:14:16 AM
"Always With Me, Always With You" by Joe Satriani:
/>
Two marches by Sousa, "Liberty Bell March"
/>
and "Stars and Stripes Forever"
/>
The MIDI files were altered for two channel by me so I included them below. I am going to upload the new boards shortly. I do not plan to mass produce them, but you can send them to board houses if you wish.
Dan
Post by: profdc9 on May 20, 2018, 03:34:01 AM
Tocatta in Fugue by JS Bach
/>
Symphony No 5, 1st movement by L Van Beethoven
/>
and my personal favorite, "Spain" by Chick Corea
/>
My edited MIDIs are attached. Thanks for watching!
Dan
Post by: Hydron on May 20, 2018, 10:58:04 PM
Post by: Mads Barnkob on May 22, 2018, 10:01:49 PM
Is it more the MIDI editing that gives the hick'ups than the pulse skipping? I think so as its only the first and last videos that show that behavior, all the classic pieces seems pretty smooth without stops.
I love 01:08 to 02:10 of Tocatta in Fogue, that is just a insane test for a DRSSTC!
How was your MMC, IGBT and primary/secondary coil temperatures after a 7 minute run like this?
Post by: profdc9 on May 23, 2018, 08:09:46 PM
I have an infrared camera that can measure temperature and I'll take a picture with it when I run it. After I run it for 5 minutes or so, it is warm to the touch, but certainly not hot. Probably 40-45 C or so. The heatsink I use is an "L" channel I cut from aluminum, with a silicone heatsink pad between the IGBTs and the L channel.
The MIDI editing does produce some hiccups, you can listen to the original MIDI if you want to hear the hiccups. I also did not do a perfect job of eliminating all of the third notes, because some of them can be hard to find, so some notes end before they should. Examining the interrupter code, I am not sure what happens if the interrupts for TIMER0 and TIMER1 fire at the same time, if a pulse is lost or not or delayed, and that causes hiccups.
My fiber optic cable is a little flaky too, I need to get a better one. Unfortunately I got the wrong kind and I am going to have to special order the right kind. I know when the fiber optic cable is very loose, the connection is intermittent and the output gets a little more staticky. But perhaps its the pulse skip too a little bit, perhaps I should run without it.
You should see the bewildered looks I get from my neighbors. But they are generally favorable.
Dan
Post by: profdc9 on May 24, 2018, 03:15:42 AM
"Little Wing" by Jimi Hendrix
/>
"Layla" by Derek and the Dominoes (I tried my best)
/>
"Take Five" by Dave Brubeck and Paul Desmond
/>
"Cliffs of Dover" by Eric Johnson
/>
I attached a zip file with the midi's in it.
Dan
Post by: Uspring on May 24, 2018, 12:09:03 PM
I was was wondering about the pulse skip. I usually tune my coils to just not hit the OCD even for long bursts. The idea is to supply max voltage and current simultaneously to the primary. The period of max input power is quite short, though. Primary current rises, goes through a max and then drops considerably again. Tuning my primary lower would increase input current (and power) during the later stage but also causes my OCD to trigger and kill the burst before that.
With pulse skipping, the OCD triggering can be avoided, so you could have possibly max input power even during the later stages of the burst. Did you observe something like this? I don't know, whether having lots of power to the end of the burst increases arc length. My coils don't show this, but it well may have been caused by the small amount of power during the end of the burst.
Post by: profdc9 on May 24, 2018, 04:22:00 PM
I have seen similar spark lengths both with and without pulse skipping, but the power available to my coil is not that great, since I currently only have 15-20 A and 120 VAC to work with, before the lousy power factor is accounted for, so perhaps only 1 kW of actual delivered power. The idea is to deliver power more continuously during the whole burst rather than only briefly at the end of the burst before the overcurrent trips. There is probably some compromise pulse length that achieves a good arc length while minimally stressing the IGBTs. Short pulse lengths can achieve good results but high peak currents shorten the life of the IGBTs. Long pulse lengths can work too, but non-interrupted SSTCs definitely underperform interrupted DRSSTCs. QCW systems have low duty cycle but very long ramp-up time (millisecond scale). The idea behind pulse skipping is that it allows one to lower the current and increase the peak current time. This strategy is probably good to a limited extent, but if the peak current is too low then the arc length is likely going to be reduced. I have not yet figured out the compromise, but the UD2.9 controller is going to help do that. I designed a new interrupter for the purpose of exploring these issues and I am waiting for the PCB to be manufactured. The new interrupter has two selectable pulse width lengths as well as bps length, and also interrupts on the peak of the AC cycle to try to improve power factor and allow burst duty cycle control, and so I should be able to try longer pulse lengths (I am currenly limited to about 160 microseconds). This interrupter is in the DRSSTC pack and you can look at the schematic if you wish.
Dan
Post by: Uspring on May 25, 2018, 12:00:00 PM
A burst length of 160us might be too short to take notice the effect of the arc limiting primary current significantly. So likely you will notice pulse skipping up to the end of the burst.
Quote
Short pulse lengths can achieve good results but high peak currents shorten the life of the IGBTs. Long pulse lengths can work too, but non-interrupted SSTCs definitely underperform interrupted DRSSTCs. QCW systems have low duty cycle but very long ramp-up time (millisecond scale).
"Underperform" depends on how you define power input. In terms of line power requirements long bursts are costly. But if you relate arc length to power during the burst, QCWs are very effective. My DRSSTC can create a 2m arc form a 320V 700A input, while Loneoceans QCW1.5 can reach almost that size (1.75m) with a 300V and 150A source.
There is a large unexplored region of intermediate length bursts nicely accessible with pulse skip technology.
Post by: Teravolt on May 25, 2018, 08:52:38 PM
Post by: profdc9 on May 25, 2018, 09:43:13 PM
It is a small modification of the UD2.7C interrupter. Whereas on the UD2.7C interrupter an overcurrent event clears the overcurrent flip-flop and can only be set again by another interrupter rising edge, the UD2.9 is selectable to that each rising edge of the current, the flip-flop is set. If the overcurrent does not reoccur, the enable flip-flop is set and the gate driver turns back on. The jumper J16 closed allows the rising edge of the current to set the flip-flop. If it is closed, the flip-flop can only be set again by the interrupter rising edge, which is the old behavior.
So no microprocessor is required, it is just a simple modification of the UD2.7C driver.
Dan
Post by: Maju on October 18, 2020, 03:28:04 PM
It's preatty old topic but I'd want to create another becouse it's directly realted to @profdc9 and his driver. This is so greate that somebodys made UD fully THT assembled with SKP feature! Thx profdc9!
But I have a reloviutionaly proposal for @profdc9. Would you be able to implement phase lead as it's done at Hammerton Predikter?
The lack of independent control of the rising and falling slopes of the phase lead is in my opinion the biggest disadvantage of all UD 2.x. I feel it clearly when driving heavy bricks when the differences in tone and toff are significant.
There are also some other advantages:
- you don't need to use adjustable coils (they are sensifive and hard to find in stores - I have to remove them from old xD radios), a simple coil for a few cents is enough
- the multiturn potentiometer adjustment is very precise and everything works very stable.
I know this is problematic because we need an additional 74HC04, but the changes are not really big.
(https://highvoltageforum.net/proxy.php?request=http%3A%2F%2Ffile%3A%2F%2F%2Fhome%2Fmateusz%2FDesktop%2Fdrsstc%2FDRSSTC-PCB-SKP-UD2.9%2Fud29C_predikter%2Fmodif.jpg&hash=8a8e1f32df60188f256cbd71c22a963545998e3e)
I tried to modify the design on my own but I totally suck at the kicad, in addition, when I try to import the modified schema to an existing PCB, I get unmissable errors (incompatible versions?) :(
UD + SKP (optional) + independent slope adjustment = Perfect Driver
It would be the best thing in my coiling life :)!
What do you say @profdc9 and you guys from Tesla society?
Cheers!
P.S. In the attachment I am sending a modified schema in kicad with this feature. I tested it with my converted aliexpres driver and it works perfectly. Probably I will write a topic with that.
Post by: profdc9 on October 18, 2020, 09:54:48 PM
https://www.ebay.com/itm/20-Pcs-Inductor-Shell-Skeleton-Empty-Ferrite-Core-No-Inductor-Coil-25-100MHZ/172915617600
You just wind your own magnet wire (I used 30 AWG) and solder it to the pins. Then there is a ferrite slug in there you can adjust the inductance with. Using this, you can make inductors with different tuning ranges.
- you don't need to use adjustable coils (they are sensifive and hard to find in stores - I have to remove them from old xD radios), a simple coil for a few cents is enough
Post by: Maju on October 18, 2020, 11:14:53 PM
The changes I've made are only between LT1016 and "AND" gates (attachement jpg).
I had a huge problems with phase lead adjust with only an inductor on my big bricks inverter. If I adjust phase lead very early (for optimal switching on), I have a problems with turning off to early and then strange things happen. That's why I have this idea and it's not only my idea. In my coutry almost everybody use Hammertons Predikter because of that issue :( But I love SKP feature, UD is easier, cheaper, more complete and more ... universal? :D
Your design is so great and it would be a pity not to use it to make this improvement.
Please analyze the changes and say what you think about it. If it's to hard, that's ok. Then the only solution will be a new board from the scratch.
Post by: Max on October 19, 2020, 12:36:38 AM
Kind regards,
Max
Post by: profdc9 on October 19, 2020, 03:39:49 AM
Dan
Those Inductor shells are also on aliexpress, that's nice to know about that. Thanks!
The changes I've made are only between LT1016 and "AND" gates (attachement jpg).
I had a huge problems with phase lead adjust with only an inductor on my big bricks inverter. If I adjust phase lead very early (for optimal switching on), I have a problems with turning off to early and then strange things happen. That's why I have this idea and it's not only my idea. In my coutry almost everybody use Hammertons Predikter because of that issue :( But I love SKP feature, UD is easier, cheaper, more complete and more ... universal? :D
Your design is so great and it would be a pity not to use it to make this improvement.
Please analyze the changes and say what you think about it. If it's to hard, that's ok. Then the only solution will be a new board from the scratch.
Post by: Netzpfuscher on October 19, 2020, 07:54:05 AM
Post by: Maju on October 19, 2020, 10:25:25 AM
@Max it's kinda funny, but turning off needs less time than turning on in my case ;) - probably because of 45nF gate input capacity and the 5.7Ohm resistors (I'm also using diodes with 1 Ohm resistor in series for turning off IGBTs). I don't want to make offtopick so I've sent image of my scope as example when I adjust "too much" lead in the traditional UD circuit.
@Netzpfuscher you still have 50% dutycycle. I think the author of this idea explains it in this documentation much better than I do :-[ :
http://www.personal.psu.edu/ahy5028/coiling/hammertone%20predikter.pdf
This idea is 100% tested by many people and works perfect with GDTs. But Predikter still has own disadvatages. It doesn't have SKP feature, uses pretty expensive and exotic comparators, only one exotic fiber optic receiver and PCB is kinda "naked". Many important components are not included in the schema. and you have to add them on external PCBs.
BTW I'm not sure if there another one common schmitt trigger is needed just next to potentiometers and diodes like in Predikter ???
Post by: Hydron on October 19, 2020, 03:54:00 PM
Note that it's a bit messy due to automated conversion from eagle or something into altium, and I think there is also a mistake in there (possibly around P5, which was meant to allow jumpering in/out of the two different phases plus the delay section). Is a long while since I built and etched it (and it's in a different country now) and I forgot what mistake was there, just that there was one that needed a bodge wire or two. Apologies if it's a bit confusing - it is for me too and I built it!
In use I found very little advantage to the fine control - didn't seem to make a lot of difference to any spikes I saw, though I was probably testing with some _very_ fast bricks (SKM200GB125Ds). I believe I ended up giving up and jumpering this section out using P5. That said, it does seem to be a more elegant solution than a variable inductor (which I also had available and ended up using) - a trimpot gives finer control and is cheaper and easier to get than a variable inductor, and I think Finn's version includes a couple more adjustable delays to add dead time if needed (with GDT output = 0V).
You can also see some options for phase lead inductor - I think it allowed me either no inductor, one or two, as I wasn't sure what I was going to be able to get. P8 also allowed the feedback and output sections to be separated - this turned out to be very handy when I was prototyping a FPGA controller as I could splice the FPGA on a dev-kit into in the middle there.
As for the GDT duty cycle question, with appropriate adjustment (of 2 or 4 trimpots!) I think you still can get 50%, either more accurately (by differentially trimming for exactly 50% duty cycle if the comparator isn't giving it directly - this is possible in my circuit) or with equal high and low periods but also with 0V dead-time periods between them (with the second two trimpots in Finn's design, if I understood it right).
I'm actually going to try and fool around with the GDT dead-time thing on the UD3 using a counter and some logic - I have some issues with hard-switching transients that I think I can maybe solve with some small C-E snubber caps and an appropriate dead-time for them to resonantly charge (giving ZVS).