QCWDRSSTC - Project Build

[ZakW]

Hello Dave,

If I'm understanding your design correctly, this supply is for gate drive power.  Using typical values for your IGBTs, gate charge is ~200nC each at 24V, so 400nC total.  30kHz * 400nC = 12mA.  Gate drive supply average current should be 12mA plus current consumed by driver chip itself, which should be available in its data sheet.

Thanks for that, I will double check my IGBTs for their specific values and run through the calculations again so I can better familiarize myself with them. I will also double check the UCC5390E datasheet.

Does your buck converter switch high side (positive supply rail) or low side (negative supply rail)?

I am planning to switch high side like in this schematic.

If switching positive rail (as is more common), then emitter voltage includes fast switching edges.  Standard line-powered isolated supplies work only if there is enough stray inductance in wiring to filter high edge slew rates.  Adding a common-mode choke to a standard supply would likely work, at least much better than relying on wiring inductance alone

That makes sense, thanks for breaking it down more.

Or as suggested, use a small supply designed for gate drive (designed to handle high slew rate edges of switching waveforms)

I agree, that is sounding like the right thing to do. I appreciate everyone's input regarding the power supply, suggesting alternatives, and being patient as I work through the advice. Glad I asked! I will take a look at the recommend alternatives and work on a new design.

If designing a continuous-use buck converter, E-cores have a disadvantage.  Wire experiences much more proximity-effect loss than for toroid cores (more magnetic field attempting to cross through wire, causing eddy current losses).  Toroid coils, especially ones with single layer windings, have much less proximity-effect loss.  Stranded wire also has a disadvantage of increased skin-depth loss.  (Unless strands are individually insulated.  In that case, called litz wire, losses are lower.)

That is good to know, I will keep that in mind. Given the increased skin-depth loss of stranded wire, would simply using 12awg solid perform better? I will likely see how the stranded wire works, but just curious if I do decide to rewind it later on.

-Zak

[Anders Mikkelsen]

One more question regarding these power supplies - the power output is so low at only 2.4W.

If I'm understanding your design correctly, this supply is for gate drive power.  Using typical values for your IGBTs, gate charge is ~200nC each at 24V, so 400nC total.  30kHz * 400nC = 12mA.  Gate drive supply average current should be 12mA plus current consumed by driver chip itself, which should be available in its data sheet.

Does your buck converter switch high side (positive supply rail) or low side (negative supply rail)?  Less common to switch negative rail, but that's what I did for my QCW buck converter.  That way IGBT emitter is tied to rectified line voltage with no high frequency switching voltage.  Standard isolated supplies work fine.  If switching positive rail (as is more common), then emitter voltage includes fast switching edges.  Standard line-powered isolated supplies work only if there is enough stray inductance in wiring to filter high edge slew rates.  Adding a common-mode choke to a standard supply would likely work, at least much better than relying on wiring inductance alone.  Or as suggested, use a small supply designed for gate drive (designed to handle high slew rate edges of switching waveforms).

That's an option in this case, given the fiber optic isolation of the buck switch driver.

Adding series inductance to limit the common mode currents through the supply Y capacitor adds an underdamped series resonance into the mix, which might be worse or better than not having it, but surely not good. Ferrites also have low enough loss at the resulting series resonant frequency (assuming a few turns through a 10 nH/n^2 core and a 4.7 nF Y capacitor) to give an underdamped response. There is also the issue of saturation of the choke due to applied volt-seconds. Residential EMI current limits into protective earth are in the tens of microamps in these frequency ranges. Putting peaks of tens of amps into earth just to save a five dollar isolated PSU seems a bit risky to me, and having switching power supplies go out of regulation from common mode transients is a problem I've had in the past, leading to death of power electronics from overvoltage.

Here is my inductor core:
    26 turns came out to 139uH (25T calculated to be 143uH)
    12awg stranded wire

I expect this will work well.  QCW use is typically low duty cycle (longer gaps between ramps than ramp time itself).  Inductor power loss is not critical.

If designing a continuous-use buck converter, E-cores have a disadvantage.  Wire experiences much more proximity-effect loss than for toroid cores (more magnetic field attempting to cross through wire, causing eddy current losses).  Toroid coils, especially ones with single layer windings, have much less proximity-effect loss.  Stranded wire also has a disadvantage of increased skin-depth loss.  (Unless strands are individually insulated.  In that case, called litz wire, losses are lower.)

E cores are worse than toroids in some respects (fringe field losses in windings, stray field EMI issues) and better in others (practical window fill factor, ease of winding), neither of which will make or break this application. Here I recommended E-cores due to the availability and low cost of a XFlux in 40 permeability. For CW type converters, toroids or edge-wound block core assemblies tend to win out, but by far the biggest factor to making dense and efficient magnetics are the core materials, and the recent advances in this field are staggering, and core shape is a secondary concern.

Stranded wire is better than solid, but much worse than litz, when it comes to eddy current losses. There's some data in this paper: http://inductor.thayerschool.org/papers/stranded.pdf

Keep in mind that the AC resistance of the wire only matters for the ripple fraction of the current, not the low-frequency ramp component. I don't think eddy current losses in the wire here will be a major issue, but litz wire is something to look into in case you have overheating of the winding.

[ZakW]

I think I have landed on some overall specs:

• Driver: Custom 1.3UD with phase lead
• Buck modulator: Asynchronous buck | dual FGH75T65SHD
• 8200uF bus cap (for now) 120vAC + doubler
• 139uH inductor + 25uF buck output cap
• Full bridge FGH75T65SHD | 200A target max current
• MMC: 10nF @ 10kV | 10S 3P | B32642B0333J
• Secondary 2.55"(64.88mm)x3.25"(82.55mm)Diameter
• Ramp generator: Fin Hammers ramp code + Arduino nano | fiber Tx

Completed:

• Secondary + topload (may have to change topload later)
• Buck inductor

Work in progress (WIP):

• Custom 1.3UD with phase lead
• Asynchronous buck driver (high side)
• Ramp generator + enclosure

Not started:

• Full bridge PCB
• MMC construction
• Buck modulator
• Primary construction | Planning to use solid 12AWG

Based on this thread (https://highvoltageforum.net/index.php?topic=2621.0) I have decided to use B32642B0333J for MMC caps as they seem to work great without excessive heating as well as the a dual FGH75T65SHD buck setup with a full bridge of the same parts. I have A LOT of them on hand so that should give me acceptable overhead.

Why do they have their MMC configured as such? Is there a load balancing benefit or does it not matter? I can't seem to find an answer online.



Why do I see some projects where a dedicated CT is used to measure primary current (also in the above picture)? If I already have CT feedback on the UD why not just probe the PCB to monitor primary current? It seemed to work just fine in my RDRSSTC.

-Zak

[davekni]

Why do they have their MMC configured as such? Is there a load balancing benefit or does it not matter? I can't seem to find an answer online.

Some MMC designs include bleed resistors across caps to keep voltage divided equally among them.  With first configuration shown only 10 resistors are required.  Second configuration needs 30.  If not using bleed resistors, only difference would be what happens if a cap fails.

Why do I see some projects where a dedicated CT is used to measure primary current (also in the above picture)? If I already have CT feedback on the UD why not just probe the PCB to monitor primary current? It seemed to work just fine in my RDRSSTC.

Measuring on UD works fine.  Measure across just the burden resistor.  Do not include phase lead inductor (ie. not directly across UD CT input if using normal UD2.7 style phase lead).

As to why others use a separate CT, hopefully someone who does that will answer.

[ZakW]

Hello Dave,

Some MMC designs include bleed resistors across caps to keep voltage divided equally among them.  With first configuration shown only 10 resistors are required.  Second configuration needs 30.  If not using bleed resistors, only difference would be what happens if a cap fails.

Thanks, I think I like the first configuration better. I will omit the bleed resistors.

Measuring on UD works fine.  Measure across just the burden resistor.  Do not include phase lead inductor (ie. not directly across UD CT input if using normal UD2.7 style phase lead).

Good to know, that is how I measured it before so I will plan to do that this time as well.