Size Comparison of Buck vs Phase Shift QCW

[T3sl4co1l]

Yeah, your layout is loose as hell, like, lashed up wires loose.  Put a ground plane under that sucker and crank everything closer together.

Alternately, use big gate resistors so it doesn't turn off faster than the layout can handle.  Deal with the switching loss somehow.

Stray inductance only cares about length or distance.  How much depends on the geometry, but it's always somewhere around ~0.6uH/m.  You can use the trace impedance to figure out how much; of course, to have a reasonably defined trace impedance, you also need a consistent geometry over a ground plane, which is why that's part of the problem.

THT transistor leads and bondwires automatically add about 10nH to the total, don't forget that either!

Tim

[Teravolt]

Hi all. I love this discussion, I assume that you have two half or whole bridges that are out of phase. from what I am understanding where the two phases intersect you are applying power to the primary. why hasn't a pulse width modulation scheme been adopted and switch on zero or shift to reduce commutation? 

[Hydron]

In theory commutation losses are minimised by always switching ON at zero current, and OFF while current is flowing. This avoids any diode reverse recovery, as the commutation from diode to IGBT only happens at zero current, with zero reverse recovery. In practice, Alan's coil isn't zero-current switching that well (part of the earlier discussion), and mine is in it's prototype stage, with sub-optimal layout etc.
If you have a look at the waveforms I posted then you can see that alternate half-bridges are phase shifted (switches each cycle), this spreads the losses out. Alan's is just phase shifting one bridge, but the driver he is using can alternate them like what can be seen in my pics.

[Uspring]

Hi Alan,

it looks to me, that the ringing you see in your scope traces are from transients on your positive DC supply voltage. The negative parts of the green and blue scope traces look fairly clean, which is due to the fact, that your scope reference is the negative rail. If you look in the middle of ramp scope shot, there are large transients on the blue scope trace, when the green trace goes down. That can only be due to the positive supply voltage wiggling.

Seems like a snubber problem in your supply.

[T3sl4co1l]

In theory commutation losses are minimised by always switching ON at zero current, and OFF while current is flowing. This avoids any diode reverse recovery, as the commutation from diode to IGBT only happens at zero current, with zero reverse recovery. In practice, Alan's coil isn't zero-current switching that well (part of the earlier discussion), and mine is in it's prototype stage, with sub-optimal layout etc.
If you have a look at the waveforms I posted then you can see that alternate half-bridges are phase shifted (switches each cycle), this spreads the losses out. Alan's is just phase shifting one bridge, but the driver he is using can alternate them like what can be seen in my pics.

It can get better than that: considerable energy is stored in junction capacitance, so you want ZVS turn-on.  ZCS turn-off is ideal but hard to arrange (operating near resonance does a good job of this, however).

Tim

[alan sailer]

Uspring,

I understand what you are saying. I'm coming to the suspicion that snubber implementation is a black art which few can master.
Currently the snubber I have on the bridge is four 2.2uF film caps mounted as close as physically possible to the IGBJTs. The buss
capacitor is a 10,000uF unit of unknown ESR. The leads from the cap to the bridge are 12AWG and about 4 inches long.

Do you think that it would help to place a 20uF film cap directly across the electrolytic?

Cheers.

[alan sailer]

So I finally got my act together and did some probing on the buck QCW. One hint to electronics experimenters,
don't buy cheap 9 volt batteries from Amazon. A box of 10 PKCELL batteries stored in the refrigerator has two
batteries left usable after 1 year.

There are two scope captures. One shows primary current (red) buck voltage (blue) and Vce (green) on the low side transistors.
Note the very low transients on the Vce compared to the phase shift.



The second shows a detailed look at the ZVS switching of the transistors. Very smooth and it looks this nice over the entire ramp.

[Teravolt]

Hi Alen nice work, in the begining you were comparing an old fashon buck and the simple driver. I asume you went with the simple driver do you have a picture of your setup? did the spikes go away when it started switching on zvs and what changed.

[Uspring]

I think you picked the right place to put the snubbers. For a 2.2uF cap, the effective impedance for the snubber at 10 MHz transient frequency is around 10mOhms. The transients you see would cause 1000s of amps in them. I don't really see, where that could come from. To check my hypothesis of DC supply wiggles, a scope shot of the supply voltage near the transistors would help.

Hydron mentioned parasitic inductances between the transistors. At 10 Mhz 0.1uH amounts to an impedance of 6 ohms, easily enough for transients of the amplitude you see. So keeping the connection short and fat there is certainly a good advice.

Putting snubbers parallel to the electrolytics probably won't help much. They are likely too far away from the IGBTs.

[Mads Barnkob]

Uspring,

I understand what you are saying. I'm coming to the suspicion that snubber implementation is a black art which few can master.
Currently the snubber I have on the bridge is four 2.2uF film caps mounted as close as physically possible to the IGBJTs. The buss
capacitor is a 10,000uF unit of unknown ESR. The leads from the cap to the bridge are 12AWG and about 4 inches long.

Do you think that it would help to place a 20uF film cap directly across the electrolytic?

Cheers.

4" leads are just a no-go in a low inductance design. It is all about large parallel and tight coupled surfaces. Just see how small differences it takes in these simulations for the inductance to change dramatically.

I forgot where I found these, but it was from a manufacturer of busbar/capacitor whole-unit integration.

[alan sailer]

Uspring,

I'll look into the wavefrom at the bridge input. I'll check it at the buss electrolytic and at the bridge input.

I did put a 20uF snubber across the electrolytic and, as you expected, no visible change.

Teravolt,

The link at the end of the first post is to a picture of the buck next to a phase shift QCW.

I could not (and still can't) attach that picture on these forums.

In a later post I showed a picture that shows the effects of adjusting the ZVS point. In the picture the ZVS was adjusted for the middle
of the ramp and you can see a dip in that area.  For some reason on the Buck circuit ZVS is static ie does not vary over the ramp. For
the phase shift is varies like crazy.

Why is a good question...

Cheers.

[alan sailer]

Mads,

Understood.

My current layout would look like I grafted an electrolytic hot dog on the side of it if I
was to make the leads short as possible. I may try a temporary lash-up to see if that
four inch length of wire is a problem.

Incidentally the wires from my buck convertor to the bridge (on the Buck QCW) are
about ten inches long. The transients on that coil are about 20 volts. The buck bridge
is also very nicely soft switched.

I'm still inclined to think that the fact that half of the bridge is being hard switched is
the biggest issue.

Cheers.

[alan sailer]

Uspring,

Attached is a scope shot of the voltage at the electrolytic cap (green) and four inches away at the input to the bridge (blue).
There are much larger transients at the bridge.

It would be interesting to compare another bridge PCB. I have a loneoceans EasyBridge that would need to be assembled and
replace the present (Alex Yuan) PCB.

Cheers.

[alan sailer]

Mads,

I ran the experiment to shorten the wires leading form the buss cap to the bridge. It made no
significant difference in the transients at the IGBJTs. The buss wires are now one inch long.

It did however, lower the transients at the input to the bridge. The blue trace in the previous
scope shot are smaller.

I am far from an expert about this but I think that inductance between the transistors (ie from
the PCB layout) is causing the issues in the phase shift bridge. This is what Hydron and Uspring
surmise. And it fits the evedence ie lowering the buss inductance has no significant effect on the
transients.

I also think that the hard switching of half the bridge is going to limit how good things can get
even with a best case bridge PCB.

I am certainly getting more interested in building up loneoceans board....

Cheers.

[Uspring]

The blue trace in your scope shot is measured at the IGBTs and so should be a measurement directly across the snubbers. There is something strange going on if you see big transients across the snubbers.

[profdc9]

I don't want to threadjack, but I have worked on a PCB layout for this

https://github.com/profdc9/DRSSTC-PCB-Pack/tree/master/full-bridge-multi

The snubbers should be placed very close to the emitter/collectors of the IGBTs I think.  In the design below, there are two snubbers that are placed in between the hi and low side of the half bridges.  The PCB layout uses almost all of the copper on the board to try to minimize inductance.

If the snubbers are very close to the emitter/collectors than there is a bit more tolerance of the connection of the bus cap.  I also made a braided connection between the bus capacitor and the PCB.  There are three pairs of positive and negative bus wires that are braided together so that the counter-flowing currents flow along the nearly the same paths to/from the PCB.  When shorting the bus connection wires together, I measured about 100 nH of inductance.

Dan

[alan sailer]

Over the past few days I built up another full bridge using Loneoceans EasyBridge PC board.
I was curious if the switching spikes got any better.

On the Buck QCW this board shows excellent behavior with very small transients. When
inspected closely the board shows great ZVS switching over the entire ramp.

The bottom line is that both Alex Yuans boards and the loneoceans board show similar spikes.
The snubbers on Alex's board are four 2.2Uf units and on loneoceans four 4.7uF.

I also shot a video showing how the phase tuning circuit on the SimpleDriver board creates a notch
of good ZVS that can be tuned across the ramp (within the range of the trimpot). If anyone is
interested I can post it on YouTube and link here.

The bottom line for me is that the phase shift QCW creates large spikes which cannot be controlled
by snubbers. I still like the size and simplicity of the phase shift QCW.

If anyone can explain why the ZVS point in a phase shift QCW varies over the ramp I'd be happy
to know. Also if anyone knows how to controll the spikes on a phase shift QCW I'd be happy find out.

Cheers.

[Uspring]

The bottom line for me is that the phase shift QCW creates large spikes which cannot be controlled
by snubbers. I still like the size and simplicity of the phase shift QCW.

Have a look at https://en.tdk-electronics.tdk.com/download/530754/480aeb04c789e45ef5bb9681513474ba/pdf-generaltechnicalinformation.pdf figure 19. At 10 Mhz frequency it shows an impedance of around 0.7 ohms for a 2.2uF capacitor. A similar diagram is also in the wikipedia article on film caps https://en.wikipedia.org/wiki/Film_capacitor#/media/File:Folko-Impedanzverl%C3%A4ufe.png

I don't know how typical these diagrams are for your species of caps, but 0.7 ohms might just be large enough to allow for the transients you are seeing. From the diagram a smaller cap value might be preferable or even better some paralleled smaller caps.

[alan sailer]

uspring,

I'm using Kemet F861 @ 4.7uF. And the graph does show ~1 ohm at 10MHz. I was asking loneoceans
about his experience with a small Phase Shift QCW and he designed a board with smaller loop inductance
for that project.

Thanks for your input. It's been interesting learning some more things. I originally started the whole QCW thing
because I had some time free while waiting for glass for plasma tubes/globes.  Normally people go
from a simple SSTC to a DRSSTC. I don't like the noise that DRSSTCs make so I went for the QCW.

In many ways it is a step to far for my modest electronics skills.

Cheers.