FAST high-side drivers/circuitry?

[mzzj]

At least my cores are bigger than needed, 37mm toroids for gate drives is sick. Bit smaller would be nice, but i dont know if it gives any advance in performance.

I have used 2wires primary and 2 secondary, not twisted but laid flat as S-P-S-P. I wanted to avoid twisting as it doesnt improve insulation. My GTD´s can probably take 8kV ac across them having some safety marginal as at some point I planned to use my ampliverter without feedback and provide safety insulation barrier with GTD´s

I also tried cheap coaxial cable for GTD,s but leakage inductance was way too high. I tought that it would give rather nice results.

[Pjotr]

Originally posted by mzzj
Currently i am using toroidal gate drive transformers in my 4kW subwoofer-inverter but GTD´s leakage inductance is limiting performance somewhat. With DIY toroid I can get sub-250nH leakage inductance witch is great compared to commercial e-core **** from coilcraft etc but i want faster....

Hi,

You can get very low leakage inductance by carefully winding it bifilar and covering the toroid completely. Have had good results with Teflon coated 28 AWG wire wrap wire. Downside is larger coupling capacitance. You can go a step further by using thin Teflon coax if you have access to such coax. The core is one side and the screening is the other side. But then you will run probably into isolation problems at that voltage.

But indeed the relatively new integrated magnetic couplers combined with high current low side drivers are a better option. You need a bootstrap construction then to power the driver or a small DC-DC converter (a simple self oscillating Royer will do) with low coupling capacitance.

Cheers

[mzzj]

Pjotr, see my previous post.

For some reason i get over double leakage with coax compared to my gdt's shown on pics.

[Joseph Hynes]

Reducing the number of turns on the toroid is another step to reduce L(leak). To cover the entire toroid with few turns, you'd need many parallel multifilars, more than current capacity alone would dictate.
I've been able to cut by 2/3 the turns by using strip-wound cores. They have super-high perm, and they're not terribly expensive (0.8 euro for a metglas 18mm toroid) and some compositions go up to 100KHz. The GDT gets a little warm in use ( amorphous materials are lossy) but keeps the leakage L low.

I am presuming you are on a low budget.
If money is no object, i'd use this combination:

-- integrated non-optical couplers: (TI) ISO150, (Analog) ADUM1100, ADUM2400.

-- integrated 2W isolated power modules: (TI) DCP021212

-- 9A mosfet drivers: (Microchip) TC4422, (TI) UCC37322

...and then you're free from all constraints. Free to create to your heart's content.

The solid state Tesla Coil community have done a lot of work with isolated drives for big MOSFETS and IGBT, you can search them too.

Joseph

[mzzj]

Originally posted by Joseph Hynes
Reducing the number of turns on the toroid is another step to reduce L(leak). To cover the entire toroid with few turns, you'd need many parallel multifilars, more than current capacity alone would dictate.
I've been able to cut by 2/3 the turns by using strip-wound cores. They have super-high perm, and they're not terribly expensive (0.8 euro for a metglas 18mm toroid) and some compositions go up to 100KHz. The GDT gets a little warm in use ( amorphous materials are lossy) but keeps the leakage L low.

I am presuming you are on a low budget.
If money is no object, i'd use this combination:

-- integrated non-optical couplers: (TI) ISO150, (Analog) ADUM1100, ADUM2400.

-- integrated 2W isolated power modules: (TI) DCP021212

-- 9A mosfet drivers: (Microchip) TC4422, (TI) UCC37322

...and then you're free from all constraints. Free to create to your heart's content.

The solid state Tesla Coil community have done a lot of work with isolated drives for big MOSFETS and IGBT, you can search them too.

Joseph

VAC would have even better cores than metglass.. but availlability and price

0.8 euro for stripwound is not bad, ferrites from my sources are around half euro.

Edit: ISO150 looks othervise nice but even optocouplers do much better in terms of dV/dt

Happy christmas!

[Joseph Hynes]

...Like I said ... go to Tesla coil fanatics for GDT issues.

Some real pearls of basic research into GDT leakage inductance:

http://thedatastream.4hv.org/gdt_prac.html

http://users.tkk.fi/~jwagner/tesla/S...es-gatedrv.htm
(this guy didn't just twist the wires... he actually braided them )

Joseph

[Joseph Hynes]

Edit: ISO150 looks othervise nice but even optocouplers do much better in terms of dV/dt

What difference would dV/dt make if the signal is buffered by a 9A FET driver? The only dV/dT the gate sees is that of the driver.

If you had the money, you'd be buying more of them than you need, then experimentally match the propagation delays, since I think timing mismatches is one of the things that results in distortion in class D.

the Analog chips are better, though.


(I'd use 2 drivers per MOSFET, in full-bridge, to drive the gate both forward and in reverse bias, to discharge Cgs faster)

Joseph

[Pjotr]

Oops sorry overlooked that mzzj, but have made wideband transformers that way that performed well up to 200 Mhz. But anyway the magnetic coupler option is a more straight way I think also no duty-cycle related problems. Have a look at the digital isolators at Analog Devices but the are more brands

Cheers

[mzzj]

Originally posted by Joseph Hynes


What difference would dV/dt make if the signal is buffered by a 9A FET driver? The only dV/dT the gate sees is that of the driver.

Joseph

I tought that dV/dt problem comes when voltage differential accross isolation barrier is changing too fast?

10kV/uS is minimum recommendation by APT for example.
If I switch 400v rail in 100ns its 4kV/us accross isolation barrier not accounting for any transients.

[mzzj]

Originally posted by Joseph Hynes
...Like I said ... go to Tesla coil fanatics for GDT issues.

Some real pearls of basic research into GDT leakage inductance:

http://thedatastream.4hv.org/gdt_prac.html


Joseph

First link has indeed some nice comparisions, but this was allready familiar for me. I havent tried screening yet because it seems bit impractical.