Axign/AX5689.... Fresh From the Bench: Improving GaN with Digital Control

[KL....]

For discussion.... Fresh From the Bench: Improving GaN with Digital Control

A Review of the Axign Class-D Audio Amplifier PWM Controller Demonstration Board
The AX5689 is an eight channel audio amplifier controller IC with digital inputs and CMOS level PWM outputs, enabling high-order digital control loops with PFFB/Post Filter Feedback. This article reviews the reference design board using GaN Systems output devices, configured as a two 500W channel BTL (2Ω) at 50V.

 

[voodooless]

92 SINAD isn't exactly SOTA. We've had this commercial Axign design perform slightly better:

'SMSL VMV A2 Review (Stereo Amplifier & DAC)'

Jun 30, 2021

This is a review and detailed measurements of the SMSL VMV A2 stereo "desktop" DAC and amplifier. It was sent to me by the company and costs US $999. VMV is the higher-end brand in SMSL lineup.

The look and feel of the A2 is similar to the latest line from SMSL:

It looks nicer though in this "widescreen" presentation. The rotary encoder feels a bit looks physically but is very responsive.

As noted, there is an integrated DAC in the A2:

I like the diagonal binding posts which gave me more room than I typically find in these smaller desktop amplifiers.

As...

• amirm
• Replies: 207
• Forum: Stereo and Multichannel Amplifier Reviews

• 

No GaN needed.

 

[pma]

For discussion.... Fresh From the Bench: Improving GaN with Digital Control

A Review of the Axign Class-D Audio Amplifier PWM Controller Demonstration Board
The AX5689 is an eight channel audio amplifier controller IC with digital inputs and CMOS level PWM outputs, enabling high-order digital control loops with PFFB/Post Filter Feedback. This article reviews the reference design board using GaN Systems output devices, configured as a two 500W channel BTL (2Ω) at 50V.

I have briefly read the published article. It is at least very interesting. I will need some more time to think it over and I promise to be back with my thoughts.

BTW, it is a pity that @SIY , similarly as @restorer-john , have resigned to contribute with their posts at ASR. The plurality of expert views is a necessary condition of moving forward.

 

[pma]

This paper by Mouton and Putzeys is a theoretical basis of the digital control loop as per post #1:

https://www.researchgate.net/publication/229001238_Digital_Control_of_a_PWM_Switching_Amplifier_with_Global_Feedback

 

[KL....]

Hi @pma, thank you :=)
Mentioned in the Testing Measurements/Writeup....

• The noise spectrum of the demo board is shown in Figure 3. There are two things of note here.
  • First, one can see the noise-shaping, which removes any issues of audible hiss even with extremely sensitive speakers. The low 1/f corner is icing on the cake. I measured under 140µV in the audio band (20Hz to 20kHz), and the white spectrum means most of even that minuscule amount that lies outside the regions of maximum human acuity.
  • Second... where’s the carrier? Normally, it shows up in the 100mV range for higher-end Class-D amplifiers, which isn’t really an issue, but the outstanding result here speaks to the effectiveness of the multiphase drive and balanced output filtering, as well as taking feedback from the load side of the filter.
• Imagine my surprise when changing the load from 10R to 2R caused NO change in the output voltage! This was true at several spot frequencies (50Hz, 200Hz, 1kHz, 5kHz, 15kHz), and is a remarkable result, predicting that the amp’s sound will remain uncolored even using speakers with wildly swinging impedance. A nice illustration of this is shown in Figure 4, which gives the level relative to 2.83V over the usual 20Hz to 20kHz frequency range for a pure 8Ω resistance and a two-way sealed box bookshelf speaker. Note the highly expanded scale. The amp’s basic frequency response has an inaudibly small 0.08dB dip in the top octave, but there was absolutely no measurable difference using a test resistor and a speaker.
• Figure 7 shows the distortion spectrum for a 4R load with 220W
• Swapping the Axign board back and forth with the Nilai DIY amps currently in my system, the sound remained unchanged — a high compliment!

 

[pma]

I have seen those comments of Stuart. I need to compare the posted output spectrum thoroughly to that of Purifi 1ET400A. Yes, carrier is almost missing, but, at first view, it seems to me that the noise shaped ultrasound noise floor is quite high, compared to conventional analog class D FB loop.

I did not like the behaviour into 8ohm//2uF load. It is my usual test and most class D amps fail to pass it.

What caught my attraction are the mentions about less dissipate power with GaNs and lower power loss at lower output power.

 

[KL....]

I have seen those comments of Stuart. I need to compare the posted output spectrum thoroughly to that of Purifi 1ET400A. Yes, carrier is almost missing, but, at first view, it seems to me that the noise shaped ultrasound noise floor is quite high, compared to conventional analog class D FB loop.

I did not like the behaviour into 8ohm//2uF load. It is my usual test and most class D amps fail to pass it.

What caught my attraction are the mentions about less dissipate power with GaNs and lower power loss at lower output power.

Thank you, yes, best to look at the positives, first, then determine/investigate if the (relative) negatives can be improved/nolonger (relatively) negative.

 

[pma]

That's fine. This digital PWM feedback is a logical next development step and the future is in this approach. I only do not want to slide to semi-marketing appraisals.

Regarding that HF noise, Stuart measures with AP and much higher resolution than I have. But, with lower bandwidth.

Axign digital loop FB idle spectrum (measured with AP) looks like this:

Purifi output idle spectrum (measured with DSO) looks like this:

The red line illustrates DSO method noise limit, however extended to 6MHz.

What is evident:

- Purifi has an individual carrier frequency line at 499kHz/-14dBV and its odd harmonics. But the HF noise floor, except for the carrier residuals, sits at the -80dBV which is the measurement limit.

- Axign seems to have carrier at 1.03MHz, very well suppressed. But the noise floor, noise shaped, is higher up to 600kHz, starting at 40kHz/-55dBV.

Which is better? I do not know. Both inaudible. Purely audio band parameters are better with Purifi.

 

[boXem]

This digital PWM feedback is a logical next development step

I'm not sure about this. It looks architecturally simpler at a first glance but:
1. The importance of the ADC in the loop is huge and I do not see ADCs evolving radically to enable better performance.
2. At identical order, a loop in the continuous domain will always perform better than one in the discrete domain.

 

[pma]

All true. On another note, who would have imagined Purifi audio parameters level when class D audio made its first steps in 1980's, 1990's?

 

[voodooless]

All true. On another note, who would have imagined Purifi audio parameters level when class D audio made its first steps in 1980's, 1990's?

The one guy that designed them all

 

[MBO]

I'm not sure about this. It looks architecturally simpler at a first glance but:
1. The importance of the ADC in the loop is huge and I do not see ADCs evolving radically to enable better performance.
2. At identical order, a loop in the continuous domain will always perform better than one in the discrete domain.

it was already clear when Sharp released its SM-SX amplifiers, there were already discussions about the quality of ADC in feedback and that it is practically of decisive importance. But apparently since then the matter has not moved from dead center

 

[Ropeburn]

Once again I see myself waiting for any actual improvement GaN brings, besides power efficiency. I haven't seen anything so far. 1 MHz switching frequency, but mediocre results. Hmmm.

 

[KL....]

This digital PWM feedback is a logical next development step and the future is in this approach.

@pma, re analog/digital PWM feedback to the Controller, was this level even attainable prior to the AX5689 chip? In this implementation it is used with GaN output but has already been used, successfully, with (mediocre) Mosfet/Chip output. Perhaps, there is some further PFFB processing that can be done prior to each ADC which improves the ADC conversion and improved digital/PWM processing? I wonder where the ADCs will evolve to in the near future?

 

[pma]

Once again I see myself waiting for any actual improvement GaN brings, besides power efficiency. I haven't seen anything so far. 1 MHz switching frequency, but mediocre results. Hmmm.

There is a key in the digital feedback loop, not in the GaN itself. Digital feedback needs higher switching rate, very short switching times, thus GaNs have to be used, as to make implementation possible. You guys have to be more patient, it also took long time for traditional class D to grow up from child’s diseases. More respect, please, in case the knowledge is missing.

 

[Ropeburn]

There is a key in the digital feedback loop, not in the GaN itself. Digital feedback needs higher switching rate, very short switching times, thus GaNs have to be used, as to make implementation possible. You guys have to be more patient, it also took long time for traditional class D to grow up from child’s diseases. More respect, please, in case the knowledge is missing.

Oh, I do realise it takes a lot of time and effort before we'll see something substantial, even without understanding the details. I have to admit to some bias here, in the face of what I perceive as overblown marketing hype before due time. It can be the next "big thing", but it isn't yet. Speaking generally of course, not specific to one company or group of developers.

Patience is indeed advised. Wild speculation:

Good old Bruno is cooking up something as we speak. In typical Purifi fashion, they're keeping it secret without making grandiose premature announcements, take their time, and in a year or three will present a finished and polished discrete design with a big bang.

That would answer what I've asked myself for a long time: why haven't Purifi or Hypex jumped on GaN, if it has so much potential? They probably did already.

 

[KL....]

There is a key in the digital feedback loop, not in the GaN itself. Digital feedback needs higher switching rate, very short switching times, thus GaNs have to be used, as to make implementation possible.

Thank you @pma, re ADC, does/could this mean that the ADC needs to be DSP supported, or similar, so that its resulting Digital output is PWM? Not one yet or is there one? The ax5689 already receives digital and translates/transcodes it to PWM, doesn't it? which involves the Digital Loop Filter, in this Application. As you have indicated, perhaps it is (simply) proximity and speed of application, although/perhaps (with PFFB/ADC) it requires/needs to be (applyed/received) faster than the Digital Loop Filter can (currently) provide/handle? Just thought/s that came to mind, for discussion :=)

 

[tmtomh]

I have briefly read the published article. It is at least very interesting. I will need some more time to think it over and I promise to be back with my thoughts.

BTW, it is a pity that @SIY , similarly as @restorer-john , have resigned to contribute with their posts at ASR. The plurality of expert views is a necessary condition of moving forward.

Have they told you (or announced in a thread I might very well have missed) that they have decided to stop posting here altogether? If so, have they said why? (Not your opinion - I'm asking if they've explicitly said something.)

 

[KL....]

Have they told you (or announced in a thread I might very well have missed) that they have decided to stop posting here altogether? If so, have they said why? (Not your opinion - I'm asking if they've explicitly said something.)

Thank you, please ask @SIY, @restorer-john, and @pma directly, via pm.
As you are, know that they are welcome to discuss this Application/Implementation here on ASR, and @pma is very welcome (and does, thank you @pma) :=)

 

[KL....]

Oh, I do realise it takes a lot of time and effort before we'll see something substantial, even without understanding the details. I have to admit to some bias here, in the face of what I perceive as overblown marketing hype before due time. It can be the next "big thing", but it isn't yet. Speaking generally of course, not specific to one company or group of developers.

Patience is indeed advised. Wild speculation:

Good old Bruno is cooking up something as we speak. In typical Purifi fashion, they're keeping it secret without making grandiose premature announcements, take their time, and in a year or three will present a finished and polished discrete design with a big bang.

That would answer what I've asked myself for a long time: why haven't Purifi or Hypex jumped on GaN, if it has so much potential? They probably did already.

Thank you @Ropeburn, know that you are welcome to discuss this Application/Implementation, strengths/weaknesses, especially improvements that can be done and how they can be done :=)