Delta-sigma class D

Two things go wrong, first having to do with dead time issues, although there are modulators that insure constant switching, and the second related to an unregulated or changing supply voltage, supplying the power being switched by the MOSFETs. Even small voltage changes would create distortion. This is why most class-D amplifiers apply their feedback around the modulation and the switching.

IR's IRS2092 is an interesting part, working to +/-100V, but they say little about the modulation oscillator, see AN-1138.

IR suggests MOSFETs they've optimized for audio class-D (quick switching, 200V and modest currents): the IRFI4019, IRFI4020, IRFI4024 or IRFI4212 (nice dual half-bridge FETs in TO-220-5 packages), or IRF6645 (dunno about its DirectFET package!).

One can get nice assembled PCBs for 200W amplifiers for $20 to $25 on eBay (made in China with U.S. parts), or a little more for higher power levels, and play with them yourself.

I was going to suggest exactly that.

The other issue with using "pure" delta-sigma modulation with a class D stage is that you end up with high switching losses. I _think_ that an answer to that is to use PWM for part of the modulation (essentially make it a multi-level output stage), and sigma-delta for the rest -- but then you end up with all sorts of issues with how fast to switch compared to the output filter bandwidth, the answers to which I don't know without doing some math.

One or two of the Fairchild 'unifet' or 'ultrafet' (FDPF) parts boast

Why doesn't IR offer discretes with similar performance to their duals?

RL

That's very good Trr on the Unifet (e.g. 500V FDPF7N50U @ 40ns) Too bad they don't make FETs with this fast Trr along with low RdsOn like the other new(ish) TO220 500V-600V FETs with less than

boB

It's my understanding, short t_rr makes terrible MOSFETs, and vice versa. I'm curious if they've actually achieved something new, or if it's just specsmanship (doing it at something much faster than 100A/us?).

Tim

I think you're right ! Designs are always a compromise anyway.

The only way that I know of to get both fast diodes AND decent FETs is to use newer technology like SiC FETs with SiC diodes. These SiC products are still way too expensive from the parts that I have found, at least for products that must compete price wise... Maybe the price will drop some if they can make more 600 volt(ish) parts rather than 1200+ volts which is higher V than I need. Then they might be affordable for mere mortals in a real product. There's that compromise again.

boB K7IQ

"Affordable for mere mortals in a real product" is egocentric speech for affordable in my product in my market.

The fact that the SiC products are on the market at all has to reflect the fact that somebody, somewhere, is buying them at an expensive price to make a product that somebody wants enough to buy at a price that makes the SiC parts affordable.

If the people who are making SiC parts work out how to make them in mass-market volume, getting the sort of yields that allow them to be sold cheap, maybe they'll find a niche in the lower-voltage market.

Nobody who posts here buys enough parts to constitute a mass market.

SiC isn't much different from Si itself, aside from withstanding a considerably higher electric field. Consequence: higher Vdss for a given Rds(on) and die area, and lower Gm per area. t_rr is almost identical (also compare the few SiC BJTs that are out there).

SiC schottky across SiC MOS (at 600 or 1200V ratings) is therefore analogous to Si schottky across Si MOS (at 60-100V ratings or so).

It works, but you'll still have better results if you can afford enough Rds(on) to prevent body diode forward bias, and just do the thing synchronously (with Si or SiC MOS alone). Preferably with just a tad of shoot-through (shorting mode commutation), to prevent the body diode being blipped and generating step recovery or something awful like that.

If you can separate the diode and FET, as with traditional (unidirectional, not synchronous) buck or boost converters, the diodes available are quite good (50ns t_rr and 1200Vrrm is typical). With capacitance lower than schottky, they can still be useful in comparison to SiC (even if 'cost is no object' and you can afford plenty of big SiC).

Tim