An Engineers road movie to an FPGA based Class-D-Amplifier

[@VEC7ORlt]

Why couldn't my university be this cool?

[@pcdimmer]

Oh, if you had asked the students in between, they would have had a different opinion :) I think some of them regretted taking part because it was a lot of work. But in the end I think everyone was happy!

[@VEC7ORlt]

@@pcdimmer Its a immense undertaking for sure, but I'd have loved to participate in something like that. My diploma work way back when was also on class D power amp, it had a fancy current sensing that used the MOSFET Rdson for protection and AVR MCU for control, ESP32 didn't even exist and FPGAs were big boys game.

[@Sextusheap]

@@VEC7ORlt Fancy current sensing indeed! Do you mind sharing your thesis?

[@Sextusheap]

@@pcdimmer I always fondly remember seasons of voluntary hardship. I don't think any one of the participants regretted the effort.

You are a great teacher! You and your students can be proud of yourselves. It is not a matter of course to use weekends for student projects.

[@ref1814]

You and your students should proud. Great project and great team work.

[@Patrick-rd9qo]

What a roller coaster - I thoroughly enjoyed your trip! And what a fantastic experience for the students, solving real-world problems to a deadline - you couldn't have a better preparation for the work-life experience developing and delivering products. Great for their CVs too.

[@pcdimmer]

Yes, that's right. Working through an explicit specification according to a strict schedule was certainly a special experience overall.

[@saumyacow4435]

Hey I love this. My final year project in Electrical Engineering was designing and building a class D amplifier and yeah, it actually worked :) No DSP or processors - this was back in 1990. Just a straightforward sampler and analogue feedback. Switched at (iirc) around 500 Khz. Also doubled as a very effective AM radio jamming device. Absolutely loved doing that :)

[@bbfoto7248]

Nice! Surprisingly, even some of the "modern" Class D car audio amplifiers being sold currently still have this problem of disrupting AM/FM radio reception due to radiated noise. :-P The first full-range Class D amplifiers I used were the tiny 100 watt @ 4 ohm monoblock car audio *Xtant 1.1i* amplifiers back in 2003 for a "high end" car stereo system that I installed into my 2001 Mini Cooper S. At 6.5"L x 5.81"W x 1.63"H , the 100 watt Xtant 1.1i Class D amplifiers were only slightly larger than the footprint of a typical plastic CD jewel case! :) I used 6 of these tiny Xtant 1.1i mono amplifiers to power a 3-way front component speaker system in the Mini Cooper S (Scanspeak Revelator 7" midwoofers, 4.5" midrange, & 1" silk dome tweeters). Independent Level adjustment, fully adjustable Crossover Network Filters, Digital Delay, and 10-band PEQ per channel for each speaker were provided by an 8-channel Sony Mobile ES XDP-4000X DSP unit with a matching Sony CDX-C90 single-DIN AM/FM/CD player "head unit" as the source. The CDX-C90 stereo head unit used a digital Toslink optical output for the CD playback which connected to the Toslink inputs on the DSP unit. A matching 10-disc CD Changer could be connected to the DSP via digital Toslink as well. The Xtant 1.1i Class D amplifiers definitely affected the stereo's Tuner Reception to some degree! But I listened primarily to CDs at the time, so I wasn't too bothered by it. :P To my ears, these were the first Class D amplifiers (of those that I heard to that point) which did not exhibit a somewhat harsh or overly bright/unpleasing treble response. And the midrange, midbass, and bass frequencies were at least as clean/articulate/full as any of the Class A or A/B amplifiers I had used up to that point as well. Before taking the time and effort to install all 6 of these tiny Class D amplifiers into my car, I tested them in my home system where they powered a pair of 2-way nOrh Marble 9.0 loudspeakers that used similar Scanspeak Revelator drivers, and I was convinced that they were up to the task. The Xtant 1.1i were listed (and confirmed via AudioPrecision measurements) as having a THD of 0.7%, and a SNR actually over -105dB! This was unheard of at the time even for some high end Class A/B home audio amplifiers. The Output Power vs THD was not the typical gentle rise towards maximum power, then rising sharply as the amplifier begins to clip. Instead, THD rises to about 75% power, then drops off before rising sharply at clipping, but while never exceeding 0.7% throughout its linear range. The output section used two N-channel Phillips devices rated at 150-watts dissipation, each switching at 1.06MHz. Power efficiency was up to 80%. If you search Online you can probably find a picture of the internal PCB layout and components. It was surprisingly simple and sparse given its performance. In more recent years I've used the BikeTronics/Vivid Amps 12vdc mobile amplifiers which incorporate the Hypex UcD & Ncore Class D modules respectively. And there are currently some multichannel GaN-FET designs from SoundDigital. In my home rig, I'm currently using a ~$1350 USD stereo Class D amplifier DIY kit amp from *DIYclassD* that is using the Hypex Nilai500 modules. For whatever reason I find that I prefer this Nilai500-based amplifier over the Purifi Eigentakt 1ET400A-based designs I've heard. ANYHOW, it's a great time to be a HiFi audio/music lover with the availability of quite a few accurate, powerful, high efficiency, and low distortion Class D (and G & H) amplifiers! Cheers!

[@welderfixer]

What a great project for EE students. Your design could be a the future of audio amplifiers even though it is far more complicated than older audiophiles like myself desire. I wish you and your students all the very best life has to offer, Kevin

[@TheRealKuif]

What a cool project and great and honest story telling. I also like that you tried to improve the result after the competition. Sometimes it's not about winning, but your own sense of achievement. So you can go to sleep knowing that you made the most out of it, even if that won't change the competition result anymore. Some well-deserved sleep ;)

[@AppliedCryogenics]

This was fascinating to watch, and it seems like a hugely rewarding endeavor for both students and instructors.

[@emielv7677]

I think you guys did really well! Im impressed by that fpga board.

[@bryanhaney7930]

I'm subscribed to a mixture of nerdy channels across spheres of mathematics, chemistry, electrical engineering, programming, additive manufacturing, photolithography, semiconductor science, etc etc... and I just want to wholeheartedly say that your approach to content is rare and unique. 🙏🙇‍♂️

[@pcdimmer]

Oh, thank you so much for your kind feedback! I hope I can keep up this kind of content :)

[@michaelashby9654]

Very cool project! Congratulations. I hope to have time to build my own. This looks like a lot of fun.

[@pcdimmer]

I will talk to my supervisor to see if we want to release the schematics and PCBs on GitHub. The power board in particular worked as desired and with the other designs you can now basically achieve reasonable values. Using a Vidor 4000 board is still possible, so that you could use this device for controlling the amp.

[@marcoroose9973]

Das ist gute Wissenschaft: auch über Misserfolge berichten. Viel Erfolg beim nächsten Mal und danke für dieses spannende Video!

[@pcdimmer]

Wir sehen es sportlich und sind froh, dass wir trotz des Zeitdrucks das Gerät so fertig bekommen haben und am Finale teilnehmen konnten. Alles in Allem war es eine prima Sache :)

[@zvotaisvfi8678]

1st place in our hearts, chris !

Very interesting and inspiring project. I used to work with FPGAs back in the days so watching your movie brought back some nice memories too.

[@jorgeisaaccuastumalpastas792]

Such a great video, it is real that this kind of projects take a lot of work but at the end everypone learned something. I'm pretty impressive that you use FPGA for Audio ampliffier, that's a usage i didn't have in mind. Gretting from Colombia.

[@pcdimmer]

It wasn't the most obvious use of an FPGA, as you can achieve more for less money with a specialized audio DSP. Thanks for the nice feedback!

[@matsstenfeldt1549]

Back in the days I owned a Sharp SM-SX200 class D amplifier for a while. It was fantastic. THD 0.05% and 200W output. The switching frequency was at 5.6 MHz! It was a total monster for its time. There where some interesting patents for the design I remember. The enclosure was solid because with the hood of it doubled as an excellent radio transmitter.😂 One of the best amplifiers I have owned. Totally clean neutral sound.

[@pcdimmer]

Even with the THD of just 0.25% achieved at the end, the device already sounded great. After my vacation I will also try upsampling to 1.5MHz or more with PDM (i.e. DSD) instead of the current PWM. This should also reduce the THD and the variable switching frequency should reduce the switching losses somewhat. We had also tested switching frequencies around 10 MHz in between. But the 100V GaN semiconductors got quite hot. EPC has even better GaN components here, but we had agreed on the GaNSystems components.

[@matsstenfeldt1549]

@@pcdimmer You and your students did a great job. I have developed embedded systems for a living. Trying to get the first iteration up to 100% spec is a waste of time. Fail fast. Fail cheap then iterate. That’s my preferred design philosophy. “Failure” also gives information. Often more than success. Most of all I do avoid paralysis by analysis. Great video! I enjoyed it very much.

[@felixyasnopolski8571]

@@pcdimmer From a viewpoint of a power electronics engineer, you absolutely don't need GaN components here :) There's plenty of mosfets rated for such voltages and controlling them would be a breeze, compared to GaN

[@Luke173]

Thanks for sharing this interesting vlog. Although I do not understand all electro technical details. Must have been a nice project & trip for your students! Gruss aus dem Süden.

[@edgarskarklins3572]

Wow...it includes so much detail, it would be a great FPGA and audio learning kit. A bit pricey, but great. Maybe with some improvements it could be badis of an open source online product design course. I would love to see other submisions if they are available somewhere.

[@pcdimmer]

Unfortunately, FPGAs are still very expensive. I spoke to the Arduino people at the "Embedded World 2024" in Nuremberg in Germany but unfortunately the Vidor 4000 is not yet being manufactured again. There is an alternative with the CYC1000, but without the ESP32 and without the additional Cortex-M0+ microcontroller. Nevertheless, the costs are significantly higher than for other controller boards. Unfortunately, I don't have the time for a proper online course at the moment. We are currently on tour with our band. However, it would certainly be an interesting thing to do.

[@abijeetrs6522]

Nice! A gem of a channel. You got my sub. 😊

[@wi_zeus6798]

Huge props to you Chris, you seem like an amazing teacher. Also I am impressed by the technical complexity of this project and the fact that it was done at least in parts by students. I am an electrical engineer and know how difficult it is to get a design first time right. Amazing work! You could check out the Zero To ASIC course / Tiny Tapeout. Maybe next time you could include a custom ASIC in your design :)

[@pcdimmer]

Thank you very much! The students did a great job, especially with the power board and the ControlBoard. They will certainly be able to take a lot from this project for their future jobs. What was the price of an ASIC photomask again? I think that's beyond our budget ;-)

[@jojoposter]

​​​​@@pcdimmer Edit: nevermind, my info was outdated. Seems that the Open shuttles have been stopped. -Free actually, if you can manage to fit into an open MPW Shuttle financed by Google. The design needs to be open source and "interesting" enough to qualify. But I think student projects are fabbed quite often.-

[@mbinacedric3566]

So goood, thank you for this great content!!!

[@mitfreundlichengrussen1234]

Thanks for sharing! A very nice view into your world.

[@Analog_isnt_dead]

Nice video! For what it’s worth, you don’t necessarily need GaN for 100s of kHz switching - I design multi-MHz power converters at my job with silicon FETs, and they are cheaper, more abundant, and a bit easier to work with than GaN. Easier to find gate drivers as well. I’d especially recommend looking into Infineon and their “OptiMOS” line of FETs

[@pcdimmer]

Yes, fast MOSFETs could have been used for the subwoofer channel in particular. But then we didn't want to make different designs and set up all channels symmetrically. Originally we had planned to go to approx. 11MHz (DSD256) - so the use of GaN seemed the most obvious choice.

[@Analog_isnt_dead]

@@pcdimmer wow 11MHz, yep that is definitely GaN territory ;-)

[@tijaelecphonics]

Sehr unterhaltsam! Super Ausstattung im Kasseler Labor. Viel erfolg beim nächsten mal.

[@pcdimmer]

Dankeschön. Da schon Fragen nach Details zur Leistungsplatine kamen, plane ich noch mal ein detaillierteres Video im Labor in Kassel. Mal sehen, was ich da machen kann. Wir haben noch viel mehr zu zeigen :)

[@ChrisgammaDE]

I'd be super stoked on a video, where you focus on the control loops of a "digital" class D amp. Most people in the field still stray away from it

[@pcdimmer]

Yep, we had problems here, too. We implemented a common PI controller with proportional and integral part. As reference-value we used the 48kHz audio-samples and upsampled them to 192kHz. To measure the output of the Class-D-Amplifier we measured the output-voltage after the LC-filter, fed it back to another Audio-ADC with 192kHz samplerate and tried to implement the control loop. But we learned, that 192kHz is not enough to bring the THD below 0.5% as the connected speaker creates some disturbances. We better had used a much faster ADC with some MHz-samplerate as with the FPGA we were able to implement the PI-controller with up to 5 MHz. Our THD could be much lower with a better signal-feedback strategy.

[@ChrisgammaDE]

@pcdimmer That super valueable feedback. Thank you!

[@bradmesserle999]

Welcome to Texas. I hope you enjoyed some Texas BBQ while on your Engineering adventures! You will knock'en dead next time!

[@MrGirgitt]

The content and production quality level of your videos is outstanding. Having a brief experience with FPGAs during my studies I can only say your team’s result is incredible given how little time you guys had to assemble, debug and integrate such a complex system. Is there a paper summarizing the competition?

[@pcdimmer]

Thank you very much for your kind feedback. Let's see, we haven't talked about further utilization since the final. There are already many very good publications in this area. But we are planning to use this amplifier device to develop a measuring device for a special type of component. There will certainly be something more to come.

[@MrGirgitt]

@@pcdimmer thank you for additional input - which brings to mind more questions but will patiently wait for a reward in a form of upcoming excellent videos :)

[@alexscarbro796]

Thank you for the detailed video. What make and model of reflow oven did you use in this project?

[@pcdimmer]

We used an older Paggen RO250BF-SF Reflow oven. But we are in the process of purchasing a new vapor phase oven (Paggen SV540). The operation should be much better - let's see :)

[@bartconinckx]

interesting that a substandard format like MP3 was used to test the audio.

[@pcdimmer]

Yes, you're right: MP3 is not the format of choice if you want to test an audio amplifier. Compression causes here a significant loss of resolution, especially in the higher frequencies (above 5kHz). However, there were more practical reasons for this: we needed a format that could be read from the SD card even with a lower SPI clock. In our first setups, we connected the SD card via jump wires and were able to clock the SD card at a maximum of 12 MHz. This meant that WAVE files could not be transferred smoothly. MP3 was simply much more relaxed. Later, we had an uncompressed audio ADC that sampled at 96 kHz and 32 bits.

[@MobiusHorizons]

I have recently gotten into HF amateur radio, and learned about the research into using switching RF amplifiers for single sideband signals, which are typically amplified with linear amplifiers in most comercial designs. I had been considering attempting to synthesize the SSB digitally in an FPGA so that it takes in i2s audio, frequency and mode (upper or lower sideband) and directly feeds the switching amplifier (in this case Class E). The inputs to the amp are a square wave representing the phase information of the signal. To modulate amplitude, the input voltage to the amp is modulated, which I was thinking of using a boost converter for, so the FPGA would output a PWM value that tracks the envelope. Obviously audio frequency has significantly different constraints than RF, but otherwise it's interesting to see a similar idea implemented successfully.

[@pcdimmer]

Yes, the applications of FPGAs are truly universal. I'm playing with the idea of building a software-defined radio (SDR) at some point. Good luck with your project!

[@dirkw4881]

Durchaus beeindruckend. Welche Abtastrate wurde bei den ADCs verwendet und mit welcher Auflösung lösen diese auf?

[@pcdimmer]

Hi, wir haben Audio-ADCs von TI verwendet. Diese haben eine technische Auflösung von 32 Bit und Tasten mit bis zu 192kHz ab. Das Audio haben wir nur mit 48kHz abgetastet, die Feedbackschleife mit 192kHz. Im FPGA habe ich das Audio gegen Ende für die Signal-Ausgabe sogar auf 1,5MHz hochgesampled. Bis dann 👋

[@dirkw4881]

Die GaN Schalter haben ja den Vorteil extrem schnell sein. Denken Sie es wäre denkbar mit einer extrem hohen Abtastrate der ADCs hier noch etwas gewinnen zu können und den Verstärker damit zu verbessern? Alternativ könnte man mit Schieberegistern und mehreren Addierern sowie einem Dividierer am Ende ja auch Oversampling probieren um die Auflösung zu erhöhen. Dadurch könnten Filter (z.b. Anti alising, ... stark profitieren?).

[@antiphlex]

I'm happy to test prototypes in a commercial music mixing/post-production environment.

[@pcdimmer]

Well, both have their advantages and disadvantages. Testing a finished device has its own appeal, as you have to make an unbiased judgment and can compare many different devices. But building a device from scratch without having done anything similar before was a great experience. I did lose hope from time to time when we couldn't program the FPGA, but such adrenaline rushes are bearable - if it doesn't happen too often :)

[@AnnaVannieuwenhuyse]

@@pcdimmer I think they were offering to test your prototype for you, in their environment as mentioned. 😅

[@PimpinBassie2]

Is it possible to 'inject' a DSD SACD 1 bit PDM stream directly into the amp?

[@pcdimmer]

Yes, this is possible: as the FPGA can handle this bitstream directly, we could use combinational logic to inject this bitstream to the amp. But to adjust the volume or to apply the filters I would suggest to use a low-pass or a sinc-filter to convert the DSD in an appropriate way. Anyway this is one of the remarkable features of an FPGA to handle both PCM or DSD.

[@HORNOMINATOR]

nice interior design 🖖

[@FabS0ag]

Dali speakers my favourite.

[@BlackenBurg]

I'm curious about the amps performance under sustained load. I know Meyer Sound Labs, in conjunction with the AES have developed newer tools like M-Noise (Like pink noise, but with much higher crest-factor) and other measurement parameters. How long can you sustain peak voltage on this amp? Great work team and thank you for sharing this journey! (from Australia)

[@pcdimmer]

Full load is indeed a challenge at the moment. Since we could only find a limited number of fixed inductors at Mouser.com or Farnell that meet our specifications (capability of beeing an inductance even at 5 MHz, no or low saturation, etc.), we used an inductor with 2.8A rated current (rather, two in series to achieve the required inductance of 44µH). At 48V and 135W, this is exactly the rated power, but it does mean that the small inductors get very hot: we measured 150°C at the inductors under full load. In the meantime, we have also operated the amplifier at 145W for shorter periods of time (even at 200W for a few seconds), but I would not leave it running at full load for long periods in its current state. The voltage itself is not a problem.

[@andreavico6198]

Can you explain the power stage more in depth? To me is more interesting than the user interface part.

[@pcdimmer]

Hi, yes, this will be a good topic for a follow-up video. In general we are using GaNSystem 100V Enhancement Mode switches with around 5 Ohm Gate resistance and 5ns of dead time for the power stage. The voltage-supply for the gate was quite challenging as our driver-ICs had a close undervoltage lockout-level that kept us busy. That was the reason for the self-made SRC for the gate-supply. I will collect some interesting points and maybe comments for a follow-up video. Best regards :)

[@leopeters1021]

@@pcdimmer Nice work! I'm also very curious. To me it sounds like the deadtime might be a noticeable THD-contributor at these switching frequencies as it leads to a current dependent disturbance. And as a consequence it would show up as harmonics of the reference waveform. Greetings from Munich. :)

[@antonioconte4254]

Hi Prof, what a great idea i hope that it could became early a product: perhaps with high grade products like PURIFI or HYPEX! Following current rpoducts around i see that topping and highend DAC products are still moving to 1 bit decoding methodologies (i think FPGA based): basically it uses SDM (DSD) format to play applying only PCM conversion if needed. DSD format is very used for highres music and any PCM could be converted (upsampled with dithering) to DSD. Using this standard you don't need a particular player and could be the best general purpose DAC/AMP on the state of the art. Please take a look at mola mola DAC internal technology.

[@pcdimmer]

Yes, you are absolutely right. We used Sigma-Delta-Modulation with up to 10MHz in this project to drive the Gallium-Nitrid-Amplifier. First, we upsampled the 192kHz Audiosamples of the Audio-ADC to above 5MHz, fed this bitstream to the digital-output of the FPGA and to the gate-drivers of the GaN-devices. We went out of time to optimize the whole system and ended up with this meager THD-value (compared to the other teams). In the end I was happy with the 0.25% THD. If we would optimize the control-loop with a better and faster ADC, we could have better results. Using this DSD signal to a pre-amp we got very good audio-signals, so this technology is really promising.

[@antonioconte4254]

@@pcdimmer There's a very appreciated software HQplayer that uses different PCM/DSD conversion strategies with a huge different dithering options... I mean to use 1bit trail from DSD to drive PWM instead comparator.... in my opinion a straight digital input amplifier (without ADC) will be a really killer application: and, in case of ADC needs, is very simple to add it externally... Looking at Hypex products like NCx500 it has 150Mhz pwd clock and output filter around 70KHz because native SDM noise repulsion.... why not to ask them? Many thanks for your great ideas sharing... please don't stop! 😄

[@stefanweilhartner4415]

sehr geiles projekt, respekt! im endeffekt ist der benchmark bei den feinen consumerkisten mit dem TPA3255 chip von Ti, wenn ich richtig liege. habt ihr da wide band gap GaN von infineon drin, die bis 1,5MHz gehen? ich hab jetzt nicht nachgesehen, wo die maximal schaltfrequenz beim TPA3255 liegt aber gefühlsmässig wohl fast eine dekade darunter. eine frage noch zu den ausgangsfiltern: kommt da noch viel HF scheiß durch? bzw. könnte man das audiosignal in den höhen mit einem filter ab 10kHz etwas boosten und dafür den analogen ausgangstiefpass in der grenzfrequenz auf 10kHz legen. die höhen sind ja vom ausgangspegel sehr gering und von der auflösung dadurch etwas begrenzter im vergleich zu den bässen. vielleicht kann man damit auch noch ein, zwei dB THD+N rausholen. ich glaub ich bin mal über eind THD messung einer TPA3255 schaltung gestoßen wo verschiedene ausgangsspulen getestet wurden. die unterschiedlichen qualitäten haben einiges gebracht. ich nehme mal an dass die sättigung des kernmaterials da eine große rolle gespielt hat. wenn ich das richtig im kopf habe, war das eine THD kurve über die leistung und da waren spulen signifikannt im vorteil, die eine grössere sättigung erlaubten. wenn ich mich recht erinnere, und das ist jetzt schon etwas fraglich, ist die sättigun nicht abrupt ab einer gewissen leistung eingetreten, sondern schleichend. aber das ist schon ein paar jährchen aus. vielleicht betrifft euch das eh nicht so stark, weil ihr vielleicht aufgrund der hohen schaltfrequenz eine höhere grenzfrequenz habt. in der messung haben die coilcraft spulen gehabt, ähnlich dem referenzdesign von TI. ich hab letztes jahr gesehen daß würth unter eigenem namen auch ähnliches hat. könnte günstiger sein. keine ahnung. die 650 bucks für euer design hat mit etwas geschockt. sind die GaN switches + treiber der fetteste anteil? ich fände es prinzipiell cool mit 32 bit digital in eine aktivbox zu fahren. mit 32bit hat man die lautstärke auch gleich dabei. und dann einen preamp mit einsteckkarten wie beim pc mit analogen eingängen, mal symmetrisch, mal unsymmetrisch, mal eine FM radiokarte, mal eine mit DAB(+), eine phono-karte, eine HDMI in/out karte die das audiosignal abgreift. eine streaming karte mit wlan oder ethernet, .... und ein 32bit I2S anschluß zu den boxen, wo dann class-d amps drin sind, ganz auf die boxen abgestimmt. gutes gelingen und weiter so! ich glaub es gibt für den technik nerd nichts cooleres als an so einem zeug herum zu basteln. und drauf geschissen auf den 6. platz. musik ist raus gekommen und der olympische gedanke zählt hier am meisten.

[@pcdimmer]

Nach dem LC-Filter ist fast kein HF-Anteil mehr dabei. Am Ende des Videos (bei 32:02) habe ich noch Oszi-Bilder gezeigt: dort habe ich die Ausgangsspannung am Lastwiderstand direkt nach dem LC-Filter mit einer 200MHz Differential-Probe gemessen. Wie man dort sieht, ist nur noch ein ganz minimaler Ripple auf der Sinuskurve zu sehen. Das Filter arbeitet somit schon sehr aggressiv und lässt keinen nennenswerten HF-Anteil mehr durch. Unser IIR-Filter für das Upsampling war dann aber etwas zu scharf eingestellt, sodass ich mit einem High-Shelf die Höhen wieder etwas anheben musste (siehe Bildschirm bei Zeit 32:29). Wir hatten bereits die Würth-Induktivitäten verwendet, die grundsätzlich recht gut für diesen Einsatz geeignet waren. Einzig der Innenwiderstand der Spulen war etwas hoch. Die rund 650US$ sind direkt die Preise, die wir bei Mouser bezahlt haben - also keine Skaleneffekte eingerechnet. Allein ein Recom DC/DC-Wandler hat 7€ gekostet - davon hatten wir am Anfang 10 Stück drauf. Ein Trafo zur Wandlung von 48V auf 15V kostete uns fast 20€. Die PCBs waren ebenfalls recht teuer, da wir lediglich geringe Stückzahlen gekauft haben. Die GaN-Halbleiter waren noch das günstigste :) Danke für die aufmunternden Worte - wir haben es am Ende genau so gesehen und waren froh darüber, dass wir im Finalen teilnehmen durften.

[@winsrrow8125]

Whats the program u used for the simulation???

[@pcdimmer]

This particular program is made by the swiss company Plexim and is called "PLECS" (www.plexim.com). It is great for simulating ideal circuits (e.g. to simulate different inverter-topologies) or to create rapid prototyping software as it can compile the project into microcontroller code. In combination with LTspice (or Qspice) for simulation with real components, it is a very powerful set of simulation-tools.

[@kabellab99]

It's bit of an off-topic, but do you think you could use an FPGA to convert the usb-a signal of a class compliment usb audio interface to spidif or AES/EBU I/O? Because It would be really nice to use my old 2 cannel interface as a spidif expansion for my new RME unit.😊

[@pcdimmer]

Well, I'm not aware of an USB-Stack for FPGAs. It will be easier to use a Soft-Core microcontroller within the FPGA. For the SAMD21 there are information about a USB audio class compliant endpoint at the Microcontroller-Forum: www.mikrocontroller.net/topic/470707 So this seems to be the "easier" way - anyway it would be a lot work. FPGAs are good for direct processing of linear protocols. But USB needs a lot of bidirectional communication. Anyway, I already thought about audio over USB or Ethernet. But as I said, it is challenging...

[@kabellab99]

Thank you for the quick response! DIY Audio over Ethernet would be a really interesting topic btw!

[@AnnaVannieuwenhuyse]

@@kabellab99 Have a look at the FreeDSP projects with high channel count. They use an Xmos processor for the task of handling USB audio class 2 compliant endpoints as well as any other possible conversions, even time keeping and clock generating!

[@ImogeneRichards]

What a great project,tremendous effort.0.2% is a decent figure. I was wondering Is the THD of 0.2% a system architecture limit? Did you simulate the amplifier architecture mathematically to get an idea of what the limits of distortion are,and where critical areas may be? It has been suggested that the sigma delta modulation design must be carefully modeled and understood mathematically to acheive very good results. The second critical part is the design of the filter used to remove hf components. Were the distortion figures you quoted,for the real reactive load,or into a resistive load? Back reaction currents will mess with the amplifier! Finally,it has been suggested that to really test for an amplifiers true distortion performance,you must use IMD testing,usually using two tones,of a particular frequency spacing,not sure what exact signal specs are,details available online for sure. I heard that Class D has now reached the level of the best class A amplifiers. Bruno Putzey has given some brilliant interviews on the subject,and is viewed as one of the best class D design Engineers in the world,if you want to listen in he's on youtube. I love the project guys well done.

[@pcdimmer]

Many thanks for the numerous questions. I achieved the THD+N of 0.25% unregulated with the HF PWM and upsampled 1.5MHz audio. The spectrum shows that we still have peaks at 2kHz, 3kHz, etc. with a 1kHz test signal. With a suitable control, these components could be significantly reduced so that a THD+N of 0.02% would certainly be possible. Unfortunately, we used audio ADCs for the feedback that allow a maximum sampling rate of 192kHz, which is significantly lower than the upsampled audio samples at 1.5MHz. We should have used a slightly more expensive ADC with a few mega samples - without looking any further, perhaps an ADS9813 with 2 MSPS. Above all, this device would have allowed +/- 12V as input voltages instead of less than 2V. The students carried out simulations in QSpice and PLECS. The challenge here was rather to simulate a realistic image of the loudspeaker. With a pure resistive load, the THD+N values were very low, but as expected, they were significantly worse with the speaker connected so that we knew that the control was mandatory. With conventional Class D amplifiers, the feedback is fed directly into the comparator input in analog form, which leads to an extremely good control. In our design, we initially had the AD conversion, which leads to a limited bandwidth and some other limitations. We had implemented a total of four different sigma-delta modulators in VHDL: first-, second- and fourth-order SDM as well as a two-stage modulator. However, testing these modulators also showed that we absolutely needed a feedback so that the system would still function stable with higher order modulators. Regarding the output filter: yes, at high switching frequencies in the range of several MHz, it is very easy to set your own filter against you. We had some problems finding a suitable inductor and getting it delivered. Unfortunately, this is also one of the reasons why we had quite high losses. All measurements shown were with a connected resistive load. Intermodulation tests (IMD) were also carried out during the electrical tests in the USA: a 1 kHz and a 1.1 kHz signal were superimposed and the spectrum examined. However, as we did not perform well here due to the problems mentioned, I did not include this in the video. I am familiar with Bruno Putzey. I also watch Paul McGowan's videos from PS Audio from time to time. It's inspiring what is technically possible. But for now, the topic is closed. We will probably set up a measuring device for components with the power amplifier, but Class D amplifiers for audio are not a research topic for us. Thank you again for your questions and the kind feedback.

[@Johnny-es9xg]

With any device that has an output that produces EMI always put low value resistors ,eg.33ohms for 3,3v and 100ohms for 5v, on your SPI and I2C bus. I have used this method with success on a controller of a 1KW fridge compressor including isolated grounds for relay output and MCU. Use switch mode PSU's for you MCU and FPGA voltage rails as they are much faster in compensating for voltage ripple and more efficient than fixed linear regulators. LC circuits are not as good as they are on paper. I would of put a fly back diode on the grounds of the speakers just as a quick fix. Slightly less power ,but would of helped you interference issues a lot more. I see you have the data lanes between ground layers which is great.

[@pcdimmer]

Yes, 33 Ohm for the SPI bus is a good adviced to get reflections under control. To fix the bug on the footprint I had chosen direct wires - not the best idea. But as we had enough space in the ESP32-S3, we did not put much efford on it to fix the SD-card reading. For the more "important" signals, we used caged PCB-tracks and kept an eye on the overall signal-conditions.

[@norbertsos4894]

what is a Electronic AC transformer?

[@pcdimmer]

Hi, a single-phase solid-state transformer is a device with an AC input, a rectification, a DC/DC conversion (for example a dual-active bridge) and a DC/AC inverter. The main-goal is to minimize the volume and weight of the whole system compared to common 50Hz/60Hz transformers. As the DC/DC converter is switching at frequencies above 50kHz the transformer can be reduced in size.

[@hrmny_]

Are you going to publish the code / pcb schematics for this project?

[@pcdimmer]

A larger part of the code is available here: github.com/xn--nding-jua/Audioplayer. This can be used to output a PDM (or DSD) and control the GaN output stage with the IIR-Filters (EQs and Crossover). Since the control loop has not yet been fully tested anyway, this is almost the current status. Only the high frequency PWM with Noiseshaper and 6-bit Sigma-Delta-Modulator is not included. I'll talk to the head of department in later August to see if we want to publish the latest version of the code and the schematics. But I'm in vacation until 19th of August, so give me some time...

[@AGISSERVER]

hello , I am trying to build a class D amplifier with switching frequency from 500khz to 2mhz. Do you have any sample circuits for me to study? There is no information in my country. Thank you to everyone who commented.

[@pcdimmer]

Well, the general circuit is that of a single-phase inverter, as used for motor applications or PV systems: four semiconductor switches, gate driver and control. Si MOSFETs are usually used for a few hundred kHz, and GaN semiconductors for the MHz range. However, I cannot provide any more information directly here. However, I am planning a KZbin video for fall or winter, where I will go into more detail about the output stage.

[@repsych]

Thanks for the video: maybe look out for the work of Takashi Kanai. He developed the S-Master digital amplification system at Sony; it nearly killed him.

[@pcdimmer]

I briefly looked up Takashi's story: the fact that even such greats of audio history faced challenges with digital amplifiers shows that it's not a simple matter. Thanks for the tip!

[@camerong4944]

I see lots of PDM processing in here. They say DSD isn’t worth it. This displays a great real world use case of 1 bit audio transport.

[@pcdimmer]

Yes, during the work on the mixing-engine and especially on the upmixing of the audio-data (PCM audio) I realized, that PCM and PDM (DSD) has both advantages and disadvantages. PCM is very nice for audio-manipulation, meaning EQing and volume-control, as you have to deal only with the individual samples. DSD would be much harder, as you first have to use a sinc-filter (low-pass). But for the output back to the real world, DSD is nice as it has a more natural stream and is easier to low-pass back to real audio-signals. As converting between both types is possible without great pain, I have to admit that both have their justification and I will continue to use DSD for analog output, but stay with PCM within the audio-processing. That's a nice compromise, isn't it? :)

[@januslast2003]

I so wanted this to succeed. But analog Class D amps have the edge because they can use feedback.

[@pcdimmer]

In the end our problem was that our feedback ADC unfortunately could also “only” work with 192kHz like the audio itself. We should have used a sigma-delta ADC here, which allows significantly higher sample rates, so that we could have sampled the signal in the MHz-range, for example - like the PDM.

[@stephanbrenner3317]

ein sehr schönes Video, danke dafür. Ich habe eine eher generelle Frage: wie nachgewiesen wurde geht es nicht ohne feedback. Klar, jeder Effekt aus der Versorgung usw. schlagen sonst direkt durch. Wenn man nun einen FPGA basierten Ansatz wählt muss doch das Ausgangssignal AD gewandelt werden, um den Fehler zu berechnen. Aus meiner Sicht hängt doch dann die erreichbare Qualität direkt von diesem AD Wandler ab? Denn jeder Fehler der dort gemacht wird ist für das System nicht von einem Class D Fehler unterscheidbar?! Wenn das so ist hat man doch gegenüber der analogen Implementierung der PWM Stufe (DA Wandlung des Musiksignals + analoge Fehlerkompensation) keinen Vorteil mehr. Oder täusche ich mich da? Generelle finde ich: lieber effiziente Lautsprecher bauen. Die paar mW die man dann daheim wirklich braucht können auch mit class A erzeugt werden. Und da hat man viele Probleme erst mal gar nicht. Oder täusche ich mich? VG Stephan

[@pcdimmer]

Ja, rückblickend betrachtet hat ein anderes Team mit einem rein analogen Ansatz deutlich bessere Ergebnisse erzielt, da die Bandbreite sehr hoch ist und nur durch den OpAmp begrenzt ist. Aber mit einer Bandbreite von 192kHz am AD-Wandler sollte man dennoch in der Lage sein, die störenden Frequenzen von 2kHz, 3kHz bishin zu 10kHz auszuregeln. Leider haben wir einen reinen Audio-ADC verwendet. Besser wäre ein Sigma-Delta-ADC gewesen, den man mit geringerem Zeitversatz auslesen hätte können. Der Vorteil des FPGA-Ansatzes ist halt, dass man sehr einfach die Verfahren ändern kann und mit dem Verstärker auch andere Sachen machen kann. Ich hatte ja einfach mit einem Update der Logik von PDM auf PWM-Verfahren gewechselt. Den Verstärker werden wir später auch als reinen DC/AC-Wechselrichter verwenden können und z.B. für Studiprojekte damit Motoren drehen lassen können - auch dreiphasig. Somit haben wir ein sehr universelles Gerät. Dennoch stimme ich Deinem Resümee bei: allein als Audio-Verstärker wäre eine rein analoge Variante für den Wettstreit vermutlich zielführender gewesen.

[@stephanbrenner3317]

@@pcdimmer herzlichen Dank für die umfassende Rückmeldung. Aus Uni Sicht kann ich es verstehen, das stimmt. Generell ist erstaunlich wie weit das Class D Thema gekommen ist. Wenn man sich beispielsweise den TPA3255 anschaut. Im direkten Höreindruck ist dieser schon verdammt gut für eine single chip Lösung. Aber es geht eben auch nichts über eine schöne 300b Endstufe :-) VG aus DD

[@3k3k3]

Next project, create a FPGA based C64 aka SID synthesizer with midi support. "Unlimited" channels and stereo output would make a product that could be not only a fascinating way to listen to old SID songs in Stereo, but also a viable commercial product :) Thanks to the MISTer all the documentation and "cores" are available :) kzbin.info/www/bejne/iJuaiaekotiJjM0

[@pcdimmer]

Sounds nice. In the GitHub-repository I've already added a sinewave DDFS block... and adding MIDI through the SAMD21 is not a big deal. Building a basic FPGA-based synth is absolutely possible. A challenge will be a solution to smoothly change between sinewave, triangle and rectangle...

[@nothingnewhere6551]

Wow. I’m not worthy.

[@davidcolver2502]

Enjoyed this. Would like it even more without the background music.

[@pcdimmer]

Well, I actually set the volume of the background music a little too loud, that's true. There is always a fine line between energetic and disruptive :)

[@izzzzzz6]

Strange design regarding cooling. What make it look like a box? Big waste of space. Could you not use some sort of Gallium nitride regulators instead of the zenner diodes? I suggest you take a look at one of the cheapest GaN power amplifier amplifiers on the market. The Admark AD60 This can output 2x 6000W RMS at 4ohms I believe. They are the cheap competitors to amps like Powersoft where Powersoft have a dsp Admark do not generally have one. I see you are designing around the audiophile hifi market which is probably your best option with this type of design. Personally I prefer to have my DSP processing separate to my amplifiers. You can simply follow the same user system for all outputs and if one amplifier breaks down you can quickly use any spare and it's ready to go. I prefer to keep 2x non computer controlled BSS dsp units at a minimum in case one of them goes down. In this case I can still run all the amps in mono. I would suggest a different layout that saves more space and would facilitate cooling if you needed or wanted to add it later. Also. I would try testing the amplifier with heavy bass music and various types of speakers to see just how well it can handle stress. Since you are more into the electronic side of audio I wouldn't expect you to have the best sound clips and cuts in your videos. But perhaps you could bring it up to 2024 with the clips you use (not the song tracks so much but the infill jingles you use) But it would also be good to hear some more fun music for the youth as after all you are teaching the youth not the boomers. Modular stacking or racking design is much more versatile as you can keep stacking / adding to your system as needed.

[@pcdimmer]

The GaNSystems switches are bottom-side cooled, which is why they are cooled through the circuit board with thermal vias. That's fine. However, there are also GaN semiconductors that are classically top-side cooled. We were aware that we were wasting space by plugging the individual boards together in the chosen variant. On the other hand, you have to bear in mind that the boards and the entire construction were designed and developed by undergraduate students. Although a colleague and I helped with the design, the boards were actually developed by the students. I think that's a great achievement. One of the other teams, for example, had minimal analog signal processing and therefore required extremely little space on the control board. Here alone, our ControlBoard is already very large at approx. 60 cm². Looking back, we got a little lost in features and neglected the optimization of the main part. We also had numerous gate supply options to choose from, but decided on the most robust one. Bootstrapping at 5 to 7 MHz was too shaky for us and DC/DC converters were too expensive. A stacked version would certainly have been the better option in terms of volume. However, we were very careful with regard to possible damage. One of the teams in previous competitions had optimized everything for volume reduction and was unfortunately unable to show anything at all in the end, as the device no longer functioned thermally due to its small size.

[@tanaseav]

This is why i hated the university. Task: reinvent the wheel but it can't be round, make it hexagonal. It was exactly the same way, make usesless bad stupid shit 😢

[@pcdimmer]

Well, I can understand the frustration. But if you're not challenged to think outside the box during your studies, you probably wouldn't come up with innovative ideas. Sometimes it really is a hurdle, especially when you see what others have already published on a certain topic. In our case, we already knew that a THD+N below 0.1% was the minimum target - if you don't know how to achieve this, it can be stressful, but it's usually a welcome mental exercise. A couple of years ago I worked in a company: there were ready-made kits for almost everything. So if you wanted to solve a problem, you just had to take certain things from it and if there were problems, there was a special department for that. That's also the reason why I'm no longer there, but at university :)

[@tanaseav]

​@@pcdimmer It's up for debate....but since i got a bit older, I started using a different approach to making shit better, and it's pretty much bang on with the innovation side: Get something that works perfectly, and try to add a feature to that shit, or improve a parameter on that without making it bad. Don't try to reinvent the wheel, it will take you a shitload of money and a shitload of time, and you will get a wheel that worse in every possible way. Try to make a wheel that doesn't slide on ice for example...or some other thing... If my teacher would have used this approach I most likely would have went a different path...but neah...it did not happen

[@stefanhertweck]

@@tanaseav If mankind had used only a "making shit better approach", we probably would still live in the analog world e.g., scratch our nails on surfaces to make sound, no grammophone -> no turntable -> no CD player -> no cloud based server music, ... disruption/thinking outside the box is an option. Stay curious ...

[@TlD-dg6ug]

26:21 you mispelled noise-gate at the bottom but not the top. 😁 Super cool though, good job.

[@ArgentumEssence]

Why 😂

[@pcdimmer]

simply because 🤷