The Wikipedia article on class D amplifiers contains the following interesting passages:
Class-D amplifiers work by generating a train of rectangular pulses of fixed amplitude but varying width and separation, or varying number per unit time, representing the amplitude variations of the analog audio input signal. The modulator clock can synchronize with an incoming digital audio signal, thus removing the necessity to convert the signal to analog. The output of the modulator is then used to gate the output transistors on and off alternately.
…
DSP-based amplifiers which generate a PWM (Pulse Width Modulation) signal directly from a digital audio signal (e. g. SPDIF) either use a counter to time the pulse length[5] or implement a digital equivalent of a triangle-based modulator. In either case, the time resolution afforded by practical clock frequencies is only a few hundredths of a switching period, which is not enough to ensure low noise. In effect, the pulse length gets quantized, resulting in quantization distortion. In both cases, negative feedback is applied inside the digital domain, forming a noise shaper which has lower noise in the audible frequency range.
My query is whether the problems of ‘practical clock frequencies’ extend up to say DSD 256, or whether HQ Player can upsample to rates that avoid those problems. If so, then are there any other issues that would prevent HQ Player from driving a class D amplifier directly, avoiding a DAC altogether ?
Is that anything different than NAD’s Direct Digital Amplification.
This is a true digital amp (not just Class D) that is computer controlled and amplifies entirely in the digital domain, converting to analogue at the speaker terminals. This gives it the shortest signal path in NAD’s history.
As far as I know, and as the article you quoted suggests, generating the signal driving the final stage of a class D amp cannot be done entirely in the digital domain. In theory, any noise-shaped 1-bit (2-level) digital signal - including DSD - could be used for that. However, unlike a “regular” DAC’s D/A stage, which drives a high-impedance input (e.g. an op-amp), class D final stage drives a low impedance load (i.e. speaker + low-pass filter), which changes the switching characteristics of the output stage. This is why negative feedback in the analog domain is needed. For example, you get one set of pulses with an 8 ohm load and another set of pulses with 4 ohm. From that perspective, a “normal” (analog-fed) class D amp is basically a 1-bit ADC dynamically optimized for whatever load it is driving. Most probably NAD M32 is no exception; I guess that’s why they call it “DirectDigital™ Feedback Amplifier”.
Those speakers sound exactly what I was thinking about. Sadly their website is a waste of time, full of marketing drivel and no mention of the Lynx model that I could find.
Note that if a speaker uses DSP internally to control the frequency response, there is no point in sending DSD to it, since it would have to be converted back to PCM.
Not if the DSP done to PCM material is done within PCM domain?
Like many HQPlayer users upsampling PCM to DSD, have been doing?
Right now HQPlayer handles (convolution) my DSP crossover (8 channels) before my 8-channel DAC that only accepts PCM192kHz.
If I had an 8-channel DSD256 DAC, nothing would change for me . Just more processing power required. Probably need an RTX 3080 to help the CPU. No DSD to PCM conversion would be necessary for PCM content (99% of all music on the planet).
The speaker I linked above has HQPlayer built-in. I would imagine they would/could let HQPlayer do the DSP crossover , just like I already do… HQPlayer is much more than just an upsampler.
It is not converted “back to PCM”, just like with HQPlayer now isn’t. If you send it 11.2 MHz rate content, and get 11.2 MHz rate output, there are no rate conversions involved, or anything like that.
For example now when I send DSD64 SACD rips to HQPlayer for playback, digital EQ (headphone or room correction) is first performed at the source rate of 2.8 MHz and then upsampled to 11.2 MHz for output.
HQPlayer has two DSP engines. PCM engine and SDM engine.
IOW, there is exactly as much point in sending things to DSP in PCM form as there is in sending things to DSP in DSD form. Treatment is equivalent in both cases.
Granted, you can tread DSD as PCM for DSP, but there are some disadvantages to this approach as far as I can tell:
Applying DSP to 64fs or 128fs requires a lot more computing than doing it at 8fs or lower.
Unless [the large amount of] quantization noise is filtered out, DSP is applied to the signal as well as the noise, which may cause problems.
Once DSP is done, it’s not DSD anymore, so if you want DSD out, you have to re-quantize. Again, that may require getting rid of the quantization noise first, to avoid further reduction in signal level.
If the source is PCM (which is the most common scenario), going to DSD before DSP seems like an unnecessary complication.
Or I could say:
Once DSP is done, it is not PCM anymore. So if you want PCM out, you have re-quantize.
Why would I do that? I perform DSP at source rate and then feed it to my oversampling and delta-sigma modulation.
I can also say:
If source is DSD, going to PCM before DSP seems like an unnecessary complication.
Since there is enough precision, it is also possible to go to DSP and then back in bit-perfect manner. But it is not the usual use case. It is even possible if you start from let’s say RedBook, convert to DSD256 and then back to RedBook.