Class D amplifier with digital inputs

[boXem]

Theoreticallly, having an integrated amp in which the digital audio signal is upscaled to 32bit @768kHz and converted to a Three level PWM to drive an GaN based full bridge power stage, where the volume control is implemented by digitally controlling the DC rail voltage of the power supply while feeding a full scale unatinuated digital signal to the PWM modulator, should be the best possible class d implementation. Too bad no such amp exists...

- why is 768 kHz better than 500 kHz?
- how do you deal with the non linearity of the PWM conversion?
- how do you deal with the non linearity of the power stage?
- how do you deal with the non linearity of the output filter?
- how do you deal with the load dependence of the output filter?
- how do you deal with the response time of the power supply?
- and so many other questions...

Since I feel a lot of creativity in this thread, here are some publications of a very creative guy who actually happens to do the best class D amplifiers of the world:
https://www.hypex.nl/p/application-notes-white-papers/
That's a good starter to understand the real issues behind class D

 

[josedgm]

Hi guys.

By no means I'm telling that a class D amplifier is a 100% digital device. All of them has to go through a signal source (digital or analog) to PWM conversion because this is the way class D amplifiers work. (like a high frequency chopper)

And also, I'm not trying to dethrone HiFi systems of several hundreds. I'm just thinking a way to have a cheaper/simpler system with fewer components and fewer external parts for the analog domain with better performance than its equivalent in terms of cost.

I'm talking at the particular case of TAS5760M, that relays on I2S inputs:

What I liked of this TAS5760M is the fact it does not require analog signal conditioning at the input stage. As you said, the feedback is required and in this case, it is done at the PWM. However, I wonder how good is this implementation and its PSSR is about 70dB. I know other amps like TAS5631B than can have up to 80 dB in PSSR

The point here is the lack of an actual DAC in the middle of the DSP and the PWM converter. And my question is, have you tried/measured something like this?

digitally controlling the DC rail voltage of the power supply while feeding a full scale unatinuated digital signal to the PWM modulator

Exactly! If the gain is set in the digital domain (like this IC does), it will low the resolution for the PWM and quantization errors will appear. Your idea is totally awesome! Saddly is very hard to implement. But what if the volume control is done at the PWM of a DC switching power supply and we keep the digital signal (and the PWM) at its full range.

 

[Trinitrond]

This is pretty close.
https://www.technics.com/us/products/reference-class/integrated-amplifier-su-r1000.html

PWM is very similar to PDM to which DSD is based on, with a 1bit 2.8mhz signal the analog conversation is a low pass filter, which is in the output stage of PWM amplifieries as well.

https://en.m.wikipedia.org/wiki/Pulse-density_modulation

Remember the definition of digital is a pulse based signal or a switched signal, with PDM(DSD) to PWM this translation is quite simple.

Digital is a marketing buzzword for a pulsing signal, in the original PCM patent there was no mentioning of digital, that came later.

https://patents.google.com/patent/US2568724

The electrical layer is pulsing, while the logical layer is binary, you can count the pulses and store them in a logical layer, but the process of storing and counting is electric.

A CD player is seen as a digital device, but the laser reading the disc is analog with RF receiver etc.

 

[EB1000]

- why is 768 kHz better than 500 kHz?
- how do you deal with the non linearity of the PWM conversion?
- how do you deal with the non linearity of the power stage?
- how do you deal with the non linearity of the output filter?
- how do you deal with the load dependence of the output filter?
- how do you deal with the response time of the power supply?
- and so many other questions...

Since I feel a lot of creativity in this thread, here are some publications of a very creative guy who actually happens to do the best class D amplifiers of the world:
https://www.hypex.nl/p/application-notes-white-papers/
That's a good starter to understand the real issues behind class D

-768kHz is a standard sampling frequency and a multiple of 48kHz. The higher the switching freq, the higher you can set the cutoff freq of the LPF. At 768kHz you may even be able to eliminate the LPF, thus increase linearity.

-PWM is a linear process with a theoretically constant gain for a frequency response. The only non-linearity in PWM is the dead-time. With high speed GaN FETs you can decrease the dead-time and improve linearity.

-At very high fc (but at least 1/10 lower than the switching freq), load dependence is drastically reduced.

-If the PSU is also operated at the same high-freq as the power-stage and it's output voltage is only varied when the user changes the volume level, then response time is no longer a problem.

 

[Trinitrond]

See also the schematica here, digital signal to PWM, volume controling the Rail voltage.

https://www.technics.com/us/products/r1/se-r1.html

 

[josedgm]

That's a good starter to understand the real issues behind class D

Thanks for the link!

This is exactly what I was talking about. Check this out:


https://www.hypex.nl/img/upload/doc/an_wp/WP_AES112BP_A_true_one-bit_power_DA_converter.pdf
No regular DAC (straight digital source to power stage)

Of course, what they present here is a more complex implementation using permutation network in a multiphase power stage.

 

[NTK]

Thanks for the link!

This is exactly what I was talking about. Check this out:


https://www.hypex.nl/img/upload/doc/an_wp/WP_AES112BP_A_true_one-bit_power_DA_converter.pdf
No regular DAC (straight digital source to power stage)

Of course, what they present here is a more complex implementation using permutation network in a multiphase power stage.

Quoting a Bruno Putzeys post in DIYAudio.
https://www.diyaudio.com/forums/cla...eeen-class-digital-amplifiers.html#post636686

...
I used to be all for digital class D at the start of my class D career as well. I made one the first working ones in 1991 for my thesis work. Mentioned in the audioholics article (which I presume you read diligently) is another minor detail. In 2000 I put together what is still by any standard world's best performing fully digitally controlled class D amplifier. Called PPDSD, it takes DSD data directly as it comes off an SACD and converts it into analog power at 0.007% THD (and with 97% efficiency at that!) with no processing other than distributing the bits across 8 power stages working in parallel. See AES preprint 5631 "A True One Bit Power D/A Converter" or US pat no. 6,803,816 for more details. I am highly proud of this work and am still waiting for someone to beat this little beast at this game (ie. without using any form of analogue error correction or feedback).
My employer is also not a very helpful argument. By management decision (and because the Chinese like it), they have been moving away from analog controlled class D and into digital for quite some time now. I have designed the digital PWM algorithms for their controllers, and have made the lowest distortion digital modulators around (THD=-160dB).
So, if you think someone else is better placed than me to make comments as to whether digital is a promising direction for class D, please name one.
...

 

[boXem]

-768kHz is a standard sampling frequency and a multiple of 48kHz. The higher the switching freq, the higher you can set the cutoff freq of the LPF. At 768kHz you may even be able to eliminate the LPF, thus increase linearity.

What about electromagnetic emissions without filter?

-PWM is a linear process with a theoretically constant gain for a frequency response. The only non-linearity in PWM is the dead-time. With high speed GaN FETs you can decrease the dead-time and improve linearity.

It is not linear. Otherwise DACs would all be PWM instead of this delta-sigma mess, they could be manufactured in CMOS and would cost 10 cent.

-At very high fc (but at least 1/10 lower than the switching freq), load dependence is drastically reduced.

76 kHz instead of 50 kHz will not be night and day. Doesn't make the filter linear anyhow.

-If the PSU is also operated at the same high-freq as the power-stage and it's output voltage is only varied when the user changes the volume level, then response time is no longer a problem.

Problem is load variation. Increasing the switching frequency will just bring new issues.

 

[boXem]

Thanks for the link!

This is exactly what I was talking about. Check this out:


https://www.hypex.nl/img/upload/doc/an_wp/WP_AES112BP_A_true_one-bit_power_DA_converter.pdf
No regular DAC (straight digital source to power stage)

Of course, what they present here is a more complex implementation using permutation network in a multiphase power stage.

The paper corresponding to your idea is more this one: https://www.hypex.nl/img/upload/doc/an_wp/WP_AES120BP_Simple_ultralow_distortion_digital_PWM.pdf

 

[voodooless]

Quoting a Bruno Putzeys post in DIYAudio.
https://www.diyaudio.com/forums/cla...eeen-class-digital-amplifiers.html#post636686

Why not also quote the last part as well:

It is suggested that you look up the following specs:
Frequency response (with different loads), SNR, THD (over frequency and power), PSRR (including modulation) and output impedance
on several analogue class D's (like UcD, Nuforce, Halcro Lyrus) and compare this to some digital counterparts (TI, DDX, D2Audio). Then look at the effort the digital ones will need to put into solving their most obvious problems (frequency response, Output impedance, PSRR, SNR, THD, and PSRR). Only PSRR is now finally being tackled by some, by adding an ADC to sense supply voltage... You'll quickly see that digital means adding complexity to address problems that analogue solves in one go, using a simple control loop.
(Of course, you can build a semi-digital control loop using an ADC to sense the output signal and execute the loop function digitally, but it quickly transpires that this does not help reduce complexity over the analogue equivalent and again adds a DAC to the equation, only in a different spot).

 

[Billy Budapest]

A CD player is seen as a digital device, but the laser reading the disc is analog with RF receiver etc.

That’s true, the actual reading of the disc is an analog process consisting of a near-infrared laser and a photodiode receiver reading pits and lands contained on the disc, with a pit producing a different voltage level than a land—but that is where the analog part of the CD ends. The non-return-to-zero, inverted (NRZI) encoding is digital—when a pit is followed by a land or a land is followed by a pit, the programmed logic reads that as a one; when a pit follows a pit or a land follows a land, the logic indicates that is a zero.

https://en.m.wikipedia.org/wiki/Non-return-to-zero

 

[Matias]

A digital amplifier already exists, but has not become famous and few people know it for whatever reason: NAD M32 with its DirectDigital technology.

https://nadelectronics.com/product/m32-directdigital-amplifier/

 

[raif71]

I have a Pioneer AV receiver that has a class D3 amp in it. Not sure why the 3 in D3. In terms of digital audio inputs, it has coax and optical in.
https://intl.pioneer-audiovisual.com/products/av_receiver/sc-1224/specification.php

 

[Gorgonzola]

The technical load of this thread got quickly too much for me shoulder, but it is my understanding that I had a 5.1 ch receiver going back at least 18 years that performed a direct PCM to PWM conversion, (though I have no idea how to controlled gain or feedback).

That was a Panasonic model SA-XRxx, (can't remember the exact model number). The unit diect 2-3 years ago for better or worse. It accepted analog input but it was my understanding then as now that they were necessarily converted to PCM before further processing.

 

[radix]

You could also take a look at this:

Sozanski, K. P., R. Strzelecki, and Z. Fedyczak. "Digital control circuit for class-D audio power amplifier." In 2001 IEEE 32nd Annual Power Electronics Specialists Conference (IEEE Cat. No. 01CH37230), vol. 2, pp. 1245-1250. IEEE, 2001.

Abstract: This paper describes a digital speaker system with a hybrid PWM digital modulator for a class-D power audio amplifier using the oversampling and noise-shaping techniques. Results of measurements of the output signal spectrum for the audio frequency band with a sinusoidal input signal are shown. The presented speaker system achieves a signal-to-noise ratio (SNR) near to 75 dB in the audio band (20 Hz to 20 kHz).

 

[Billy Budapest]

A digital amplifier already exists, but has not become famous and few people know it for whatever reason: NAD M32 with its DirectDigital technology.

https://nadelectronics.com/product/m32-directdigital-amplifier/

Actually, NAD’s “direct digital” topology was very well known—they had a number of products based on it including the 3020 reissue, all utilizing a closed loop digital feedback technology licensed from Zetec Semiconductor—but the topology’s measured performance was worse than traditional class D and they phased it out. Wadia, too, sold a PowerDAC utilizing a similar concept. The M32 represents a second go at it for NAD using technology licensed from a different partner.

I haven’t read this (yet) but NAD explains their “Direct Digital” technology in the white paper here:

https://nadelectronics.com/wp-content/uploads/2017/03/DirectDigital_WhitePaper.pdf

 

[voodooless]

There is at least one all digital class D solution that can deliver high performance: https://www.axign.nl/

Amir tested and of these a while ago. It’s pretty good!

 

[dshreter]

So, if you think someone else is better placed than me to make comments as to whether digital is a promising direction for class D, please name one.

What a quote by Bruno. I guess nothing new should be invented by someone else ever.

I don’t doubt that a state of the art analogue implementation is better than what has been achieved by the digital approaches so far. But they do exist, the Qualcomm chip inside Sonos Amp is another example. Whether the approach over time will offer advantages in fidelity, cost, efficiency, geometry… who knows. Some manufacturers clearly see benefits. But at this point in time the best amps are not designed this way.

 

[radix]

What a quote by Bruno. I guess nothing new should be invented by someone else ever

AFAIK, Bruno did not invent digital control of class D. That's been around since the early 90s. Maybe earlier. I think the quote from Bruno is pretty much on the money -- he really does know what he's talking about.

I don't see why getting rid of the DA is so important. You have a DA one way or another. In a standard class D with digital control, you have a PWM, which is a pretty bad DA converter. In modern designs, they are using delta-sigma control of the switching power supply -- i.e., a better DA converter.

The real question is if it is better to DA to line level, chop it to PWM, and amplify that, or DA to PWM/delta-sigma/etc. and amplify that.

For example:

Matamura, Atsushi, Naoaki Nishimura, Preston Birdsong, Abhishek Bandyopadhyay, Adam Spirer, Mariana Markova, and Shaolong Liu. "31.1 An 82mW ΔΣ-Based Filter-Less Class-D Headphone Amplifier with-93dB THD+ N, 113dB SNR and 93% Efficiency." In 2021 IEEE International Solid-State Circuits Conference (ISSCC), vol. 64, pp. 432-434. IEEE, 2021.

Skimming that paper, DA to delta-sigma to switching amplification is not simple.

Marc

 

[mk5566]

I'm also thinking about this path and found it has been done by some manufacturers like Edifier, there are chips made by TI use direct PCM to PWM and amplify this PWM to drive speaker, I do like this concept very much but I believe there are some difficulties to make this work well, their newer product still use typical DAC to class D amp method, and the result is better, I think those conversions are easier to do well.
Here's a document from TI talk about this