Blast from the Past...

[JonMarsh]

Revisiting this topic and paper came about because of some former colleagues asking questions about early work that was topic related to another former colleague, who joined us from a company very involved in Class D (Extron) and went back to them, and the curious tale of how I knew their research director long before this employee did, because of an AES presentation I gave and an invited paper that resulted from that presentation. Of course, it's about Class D design and development strategies.

I shared that with Sven this weekend, as he needed reading material on his flight back from India (I don't think this is what he was expecting, though!) and he suggested I create a post for it in case anyone else might be curious...

Keep in mind, this work was done WAY BACK before a lot of things we have and take for granted, including high speed computers and modern simulation tools and device models. All the simulation and development work was done on my personal home built 80386, which was blazingly fast and capable (32 bit!!) compared with what we had in the office for Application Engineers.

And I'll admit that it may seem a bit cheesy that I cited myself (in published papers) as a reference 5 times out of 19 cited references. But hey, prior art and all!

This presentation and publication got me a lot of interesting questions from the handful of other people active in class D development efforts. Sort of like making your own invitations...

The most curious and intriguing thing about re-reading this was noting in the power supply section (SMPS, not big transformers and caps) I had been developing what I called a ZVS-QRC, or Zero Voltage Switching Quasi-Resonant Converter, and I'll be dipped if that wasn't pretty much an LLC converter, using a special IGBT with fast switching and very low tail current losses, made lower with a half sine current profile this converter used. I don't recall LLC being a "thing" back in those days, not until the late 90s, though. And LLC analysis and design synthesis is what I focused much of my work on the last 5 years. My last work was an APEC professional education seminar on that topic, about 130 foils... drop me a PM if you have Apple messaging and I can send you the 50MB file. ;^)

OK, Peabody set the wayback machine to 1990!

Page 1.

[JonMarsh]

Page 2, Class D

[JonMarsh]

Page 3, Class D

[JonMarsh]

Page 4, Class D.

[JonMarsh]

Page 5 Class D

[JonMarsh]

Page 6 Class D

[JonMarsh]

Page 7 Class D paper.

[JonMarsh]

Class D Paper Page 8

[JonMarsh]

Class D Paper Page 9

[JonMarsh]

Class D Paper page 10

[JonMarsh]

Class D paper page 11

[JonMarsh]

Class D Paper page 12

[theSven]

Awesome share! Thanks for posting this JonMarsh!

[letiennam]

Awesome share! Thanks for posting this

[JonMarsh]

You're quite welcome- it's interesting to see how much has changed, and how similar some things still are, from over 30 years ago...

[dar47]

Just saw this interesting read and I wonder if there is some more work to do on the power supply side for class D.

I had no issue with switch mode power supplies on the pro-audio side of amps for subwoofers but starting with Ncore and a nameless power supply used for these and the new Purify amps seams to not last very long under light conditions. In my very little circle of 3 people we have had 3 flaky power supplies. After my NC500 build flaked I wanted to try Purify so I went with an affordable NAD C298 with Purify and NAD's power supply and liked the sound upgrade from Ncore but worried the long term reliability of the power supply might not be there. Now I think class A/B manufacturers have had to adjust pricing from the introduction of these new wonderful affordable Class D's to the market but do they have issues with longevity of the power supplies? The power supply is complicated, is this the issue?
I started thinking just go with the newer Bryston 3B3 or 4B3, 20 Year warranty and sounds wonderful! This amp is close to $8,000 cad. but is on the lower end of really good A/b's or get the Masters NAD M23 with a little different design and hope my light use will last me.



New M23


On another note maybe I am just a worry wort but I also change the power supply in my Eversolo DMP-6 streamer to a leaner one and maybe I here a difference LOL.

[dar47]

With the new M23

[JonMarsh]

Well, you know, that's the funny thing... a standard linear power supply is pretty much rock solid, lasts dang near forever, but as is hinted in the post above, a switch mode power supply can be a bit of a tricky beast. I was a "Distinguished Engineer" the last 8 years in my career of 35 years with a German semiconductor company, and that was due to contributions in understanding how best to use different types of switching power transistors, and a much more thorough examination of some key topologies like LLC push-pull resonant converters. The popular method of modeling the latter is called FHA, for First Harmonic Approximation, which is basically treating the power topology as if it was a linear network like a loudspeaker crossover network. That works OK EXACTLY at the resonant frequency, but is various forms of screwed up anywhere else. What I showed in this paper above and described as a ZVS-QRC converter (with our special IGBTs) is basically a subset of the LLC converter. In their 2001 paper, Lazar and Martinelli proposed a discrete time based analysis mode, which I guess folks weren't keen on because the math was hard!

🤣

But that is what works... it took a training campaign internally with webinars back in 2016 to start convincing at least a few folks at my former employer about that, but it grew in scope to where the last project I did was a 120 page APEC PES (Professional Education Seminar) for 2021, which my partner in crime on this topic would have to present with out me, as I retired the end of 2020...

Now, I am still thinking about an SMPS solution for Class D amps, I'm even attending a 9 hour development forum from Europe from my former employer on Widebandgap power transistors on April 16, from 12 AM to 9AM. But I've already done some homework for that, and fired in some relevant questions and suggestions ahead of time, so it will be interesting to see how it goes. I know about half the guys presenting personally, and all on the SiC transistor side, and didn't ask easy or comfortable questions.

OTOH, Infineon has just introduced their 2nd generation 650V SiC MOSFETs, and they're doozies- and priced very competitively also. I did a comparison of a number of current MOSFETs in a simple chart, not focused on ID or Ron rating, but selecting parts with a common Qoss range and Coss at 400V characteristics- this is a key spec for resonance and high frequency switching, correlates with the net drain active area, and it's interesting to see how the other stuff falls in place from various competitors.

Look at the bottom selection area for Qoss, Coss, and Co(tr) (Time related) and Co(er) (Energy related)- commonality was the point, as these would all be usable in the same resonant switching circuit. But the conduction characteristics- the other side of the figure of merit expression, you could make it Ron X Qoss, or 1/ID X Qoss, the smaller the number, the better. Hints about dynamic switching characteristics can be seen comparing gate charge, and especially focusing on gate to drain switching charge, as this is the gate energy and correponsing time necessary for a turn-on drain voltage transition.

This is from a more recent "educational" presentation I put together for a former colleague to bring him up to speed in a new role he has... at another company.




This is all from public data sheets- no proprietary information involved. I do have knowledge of a lot of the changes in device structure and manufacturing techniques that resulted in the improvements in the just released early this year IMBG65R040M2 (M2 = 2nd generation) and some bits you can find in published paper, like the use of an asymmetric trench gate technology to both reduce voltage stress on the gate oxide and support a trench gate design achieving higher mobility and lower net source Ron (just compare the Id-Pulse ratings to see the impact from this as well as the Ron of this part versus the IMBG65R072M1H).

Fun stuff!