Friday 3 May 2013

“Pure” PCM and “Pure” DSD

I wrote a little while ago about the relationship between DSD and PCM – how DSD is a specific implementation of SDM (Sigma-Delta Modulation), and how both ADCs and DACs for PCM are built around SDM engines.  I also wrote about the algorithms that convert data between the two formats, and how the conversions are not entirely lossless.

The picture I left you with was that – for all practical purposes – there is no such thing as “Pure” PCM.  Any PCM music data is derived from SDM at some point in its creation, and has therefore undergone at least one conversion sequence.

I want to ramble further on DSD, and whether music stored and replayed in the DSD format can be any more “Pure”.  The problem is the inescapable fact that – until somebody comes up with a truly significant breakthrough – you cannot “edit” music in the SDM domain.  This has huge ramifications for the recording industry, where recording, mixing, and mastering can involve very profound manipulation of the music.  Indeed something as simple as volume control – a fade-out, for example – cannot be done in the SDM format.  And the recording industry routinely employs way more elaborate effects that would make your hair curl (have you noticed how many recording artists have unnaturally curly hair?...).

To my knowledge, there are only two studio-grade recording desks out there capable of producing commercial DSD recordings – Sonoma and Pyramix.  Of the two, Sonoma is the oldest, and least functional.  Sonoma drops the signal out to PCM for fade-in and fade-out, but apart from that offers no sound manipulation capability.  Pyramix is modern and quite progressive, but it does all its mixing in “DXD” which is 24-bit 352.8kHz PCM, so all Pyramix DSD is derived from what are essentially DXD masters.

Is there any such thing as a pure DSD recording?  Well, yes, there is, but you have to restrict yourself to transcriptions of analog tape, where no further audio processing is required.  To be fair, there is a fair amount of archival material out there which could benefit greatly from transcription to DSD for re-release.  But, for new recordings, you would need to record to analog tape and mix on an analog deck if you wanted to create true DSD recordings.  Some boutique studios do follow this approach.

And as far as it goes, that sounds all well and good.  But then I came across an interesting paragraph in a 10-year old technical paper from Philips in Holland (who, together with Sony, were the driving force behind SACD).  Here they talk about the typical “DAC” configuration used in a SACD player, and I was rather surprised to read it.  According to this paper, the low-pass analog filters that are required to convert pure DSD to analog do not possess the impulse response characteristics they consider to be necessary for high-end audio performance.  However, digital low-pass filters are more than up to the task.  Therefore, the first thing the DAC does is to use a digital low-pass filter to convert the DSD to a 2.822MHz multi-bit PCM signal.  This PCM signal is then fed into a SDM to generate a multi-bit (typically between 3 and 5 bits) SDM signal at 5.6MHz or even 11.3MHz.  This multi-bit SDM can finally be passed through a low-pass analog filter without having to sacrifice the impulse response characteristic.

So, who’d-a thunk it?  DSD gets converted to PCM and back again in the DAC of a SACD player!  It would be interesting to find out whether modern DSD DACs utilize a similar approach.  If so, then arguably, as well as there being no such thing as “Pure” PCM, there could be no such thing as “Pure” DSD either!