After 7 Years of Babyface and now some 3 Weeks with the excellent new ADI2 DAC FS I would like to request a "PCM Direct" feature, for a perfect sound, possibly even better than what now only possible with DSD.
I guess that DSP processing is still active within the audio chain when playing back any PCM input (USB), even with the highest sample rates while all DSP settings are deactivated (set to default for all 3 outputs): the only significant SQ improvement is to convert PCM to DSD, the differences between any two PCM or between two DSD playbacks (like DSD vs. DSD direct or DSD64 vs. 256) are definitive there but very subtle in comparison to PCM vs. DSD.
When DSD is converted back just for volume control (and than back internally again into bitstream fomat for sigma-delta modulation), playing back PCM directly should theoretically produce a better sound without redundant steps of PCM->DSD (for input) + DSD->PCM (within the DAC). Despite of this DSD playback is dramatically and consistently better than with any PCM input. The differences DSD/DSD direct, DSD sample rate, PCM Sample rates are almost negligible.
I have discussed this with the creator of HQPlayer, he meant that for comparison we could play back PCM converted to highest 705.6k or 768k sample rates before output to the DAC. Then due to the performance limits of the DSP would most of the DSP processing be switched off. This was not a perfect comparison within HQPlayer though, being the DSD conversion the result of 256* digital filtering and 16* of the PCM within HQPlayer .
But then I was able to make a perfect direct comparison with JRiver (upsampled to 7xx PCM) and then output it by defining foo_asio_dsd as the output device. These are obviously of lesser quality components than HQPlayer4, but the results confirm strikingly my observations: it's only the DSD input format which gives the near-optimum playback, even with this not highest quality conversion.
The only conceivable explanation I can see is some un-necessary processing by the DAC DSP. Why couldn't at least some special format PCM input get treated like the PCM which is converted back from DSD just for volume control?
Perhaps as an additional option (like DSD direct, but with volume control) or as the default behavior with some sweet-point, high rate, 32 bit PCM formats, where all of the DSP processing could be circumvented.
Tested with some demanding HiRes (96/24) files (bit depth being always the major factor), but also CD files as MQA and CD quality streamings from Tidal. Environment: Windows10. (Interestingly, there is a marked difference between ASIO and WASAPI as well, WASAPI(event) being of the markedly richer sound quality. Was even with the Babyface concistently the same experience. )
To my own qualification: I have been invited for a decade for the inner beta tests for world leading SW instruments (both sampled & physical modeling), I am a mathematician by profession with some classical musicology background.