[dreamcompiler]

The crossover distortion seen here suggests an analog Class-B output stage and that surprises me, because a digital output stage would be much more efficient. Class-D in other words. I've built digital inverters using IGBTs that produced an output sinusoidal power wave with lower distortion than the mains power. Granted these were one-offs and probably not cheap enough for production, but modern IGBTs and MOSFETS should be cheap enough nowadays that medium-priced UPSes could just use Class-D as the default solution.

Assuming you really need a sinewave at the output at all. DC output UPSes are the most efficient way to go if you can bypass the switched-mode power supply at the input of your equipment. Which most equipment has these days unless AC motors are involved.

[exmadscientist]

Phase-shifted full bridge is the way to go. (It might have another name in this area of power electronics, these things do have lots of names....)

We did a "big" inverter design a while back (500 VA was big for us; perhaps not for you). The guy who did the concept architecture suggested a PSFB design. He then quit to take a a great offer from a startup. Not really being a power electronics team, we hired a specialist consultant. The first consultant did... honestly, I don't know what he did. But it was weird. (This was a problem.) It wasn't a PSFB anymore. It also didn't work. The design then went through five more lead engineers and two more consultants, plus one more if you count me on the side watching and occasionally pitching in (I was the sister subsystem lead). It ended up being a full digitally programmable bridge and we had to figure out how to switch it. Guess how it ended up working?

Phase-shifted full bridge. Just like the first guy (and I!) said it should have been all along!

[exmadscientist]

Also I probably should have actually addressed your Class B comment:

No, they're not Class B. It's all digital PWM stuff inside. But the duty cycle gets tiny near zero cross, there's very little power in the waveform there, and there's overhead to have a switching device on at all (this is much more noticeable for IGBTs).

So it ends up being a massive simplification to just not care about that section. And it's a simplification that works pretty great, so people do it!

We had to get this truly right in the inverter I mentioned in sibling comment (as it wasn't a grid-feed or backup inverter, it was doing Something Else™ *) and just that piece was actually way harder than the entire rest of the waveform output design.

* hopefully NDA-OK spoiler: let's just say I know way, way more than I'd like to about what's inside that Chroma 61507 mentioned in the article.

[dreamcompiler]

I don't remember the details on mine but I might have had the luxury of letting the entire output voltage be positive. Hence no crossover distortion.