I would love it if you could explain the audible advantages to me, as a listener, of your software. Something in rather simple English would be preferred. If you could point me to any tests or measurements that would confirm those advantages, it would be greatly appreciated. I apologize for not being able to understand the advantages from your highly technical answers.But then you are not staying in native 24-bit! It is not native 24-bit processing anymore!
Windows audio pipeline uses shoggy 32-bit floating point just like macOS CoreAudio too.
HQPlayer uses 64-bit, 80-bit, 128-bit or arbitrary precision math for PCM as necessary.
If you start with 24-bit file, do processing and return to 24-bit it is no different.
OTOH, when your re-modulator is better than the original one, with lower noise floor, you are not adding any noise. It is just like taking 24-bit PCM, processing it and doing 32-bit output. Or if you for example upsample while doing the processing, you gain both increased dynamic range and increased banwdith. Just like with PCM too.
Then if we consider that even at DSD64 you can have ~180 dB dynamic range over the audio band, this is 36 dB more than 24-bit PCM has. If we then consider worst case 3 dB loss due to remodulation, we can see that you could do 12 remodulations before the noise floor reaches that of 24-bit PCM. If we then cosider that maximum acoustic SPL of recording would be 120 dB(SPL) with 30 dB(SPL) acoustic background noise (which is very quiet) we conclude that the recording would have 90 dB dynamic range. This means we have 90 dB of headroom in DSD64 and could afford 30 reductions of 3 dB before we reach the analog/acoustic noise floor.
At higher DSD rates we can reach 192 dB or more dynamic range. So we have even more headroom.
Since we are not stupid, we combine all DSP processing into one re-modulation step. Replay gain, speaker placement processing, digital room/headphone correction, headphone cross-feed, etc.
So in the end, we stay tens of dB below noise floors of analog recording or DAC performance.
Sure, a non-problem. We have all the nice GPUs and CPUs of today. So we can keep making better and better algorithms.
Just to be clear, I generally dislike use for marketing terms. Mixing PCM (non-marketing) and DSD (marketing) terms is disingenuous. So HQPlayer outputs SDM. And it happens to be compatible DSD DACs. It can also produce up to 257-level multi-level SDM if you like.
If you process DSD256 source to 44.1k x256 SDM output, there are no rate conversions whatsoever involved. Output is re-modulated with the modulator you have chosen, along with your chosen processing (EQ, mixing, etc), this is clear from the documentation. If you process DSD64 to 44.1k x256 SDM output, there is direct upsampling from 2.8224 MHz to 11.2896 MHz along with all your chosen DSP processing. No intermediate rates between.
As I explained before, as simplified example, if you mix two DSD streams together, without level adjustment, result is 3-level signal, equivalent of 1.585-bit binary data. So your output data may be "11", "10", "01" or "00" of which the two middle ones encode the same level, so total three levels. Or essentially a dual-bit stream.
This is much much better than the usual PCM exercise where you first have SDM output from the ADC, then converted to low PCM rates by the on-chip converter (not so great). Then you perform all kinds of processing on the PCM data you got. And then you send it to the DAC when it is again converted back to MHz rates using (poor) on-chip converter and modulated back to SDM using (poor) on-chip modulator. Totally unnecessary conversions with tiny cheap resource constrained on-chip DSPs.
Thanks.