Sunday, 4 May 2014

PCM vs DSD … again

Particularly with Digital Audio, you often run into situations where logic appears to dictate one thing, but experience favours another.  As a consumer there is little to be gained by pursing the logical solution in the face of a better sounding practical alternative.  The only reason for listening to music is to enjoy the experience.  Why choose to listen to something you enjoy less because some arcane explanation says that it ought to actually sound better?

However, for those of us involved in this business on a professional level, we cannot afford to take such a simplistic approach.  If something that should sound better actually ends up sounding worse, it is usually in your best professional interests to find out why.  In most circumstances this is far easier said than done.  Sometimes it can seem an impossible thing to get done.   In many instances the cost of doing the necessary work may be the limiting factor, and that is a very valid consideration for most people in the audio business.  In the software business, it is often a more manageable task.

Take the question of DSD playback vs playback of a PCM conversion of the same DSD data.  It is a known fact that you cannot convert losslessly, back and forth, between DSD and PCM.  DSD stores some information which cannot be adequately represented by PCM, and vice versa.  Simple information theory therefore states that if you make a PCM version of a DSD track, no matter how advanced the quality of the conversion, some information is going to be lost along the way (I will come back to this later, but for the time being we will stick with it).  If information is being lost, then the result is, by definition, a poorer quality representation.  This is an inescapable fact.  It may be, of course, that the information that is lost is information which contributes adversely to the enjoyment of the playback experience, but it is lost information regardless.  And if we were able to identify it as such, we could take steps to eliminate it from the original recording in the first place.  No, the fact remains, a PCM conversion made from a DSD original, is - de facto - a retrograde step.

So what happens if the PCM version is reported to sound better?  Do we simply argue that it can’t possibly sound better because it is inherently inferior?  No, of course not.  There may be a perfectly good explanation for it that we have not yet stumbled across.

Don’t forget that we never actually listen to digital files.  We listen to analog signals which a DAC has created using the digital files as its source material.  And there is a rather serious amount of processing that goes on between the digital data being sent to the DAC and the analog signal being transmitted out of it.  There is no doubt that this digital processing has the capability to alter the signal substantially enough to account for audible differences.

I have mentioned before that we have ordered a PS Audio DirectSream DAC as our new reference DAC.  I can’t wait to receive it.  Paul McGowan of PS Audio has been using BitPerfect and DSD Master to play his DSD content through the DirectStream.  The Hybrid-DSD files enable him to switch easily between the original DSD content and the PCM conversions for all his DSD music.  Paul reports the surprising observation that he tends to have a slight preference for the PCM conversions over the DSD originals in his system.  I will be doing the same comparisons myself when my own DirectStream arrives.

This would be an example of a circumstance where I would say that the DSD original ought to be “fundamentally” superior to the PCM copy.  Or at the very least as good as the PCM copy.  Having written DSD Master (which was used to do Paul’s conversions), I would say with some certainty that there is no basis for supposing that our PCM conversions could actually improve upon the DSD original.  Assuming that Paul’s observations are accurate and reproducible - and I have no reason to suppose otherwise - this would be a perfect example of a situation where one has little professional alternative but to try and figure out what is behind this unexpected result.  For consumers, though, it perhaps provides another - and quite unexcited - reason to go out and purchase DSD Master!!

I said earlier that I would elaborate upon whether or not data must perforce be lost in a conversion between DSD and PCM.  The obvious caveat here is that, clearly, the DSD must not have been sourced from a PCM master!  There are some glaring real-world examples here.  Some SACDs, for example, have been publicly exposed as having been mastered from 16-bit 44.1kHz CD masters!  The DSD64 format is perfectly capable of storing all of the information that is held in a red book recording.  Therefore, those SACDs, at best, have no possibility of ever sounding better than a perfectly-played CD.  Many other DSDs are sourced from Hi-Rez PCM masters.  Generally, if you were to convert such a DSD file to Hi-Rez PCM using DSD Master, it is quite likely that the resultant PCM may well sound close to indistinguishable from the DSD version.  Still and all, I don't see why it should sound better