[linux-audio-dev] Direct Stream Digital / Pulse Density Modulation musing/questions

[Dustin Barlow]

I read an interesting article on Direct Stream Digital (DSD) / Pulse Density 
Modulation (PDM) entitled "A Better Mousetrap" by Brian Smithers in the May 
2003 issue of Electronic Musician.  Since, Brian did a good job explaining 
PDM/DSD in quasi-layman terms, I'll just quote snippets from his article to 
set the stage for my questions.

<quote a="EM: A Better Mousetrap" p="96">
In DSD, the amplitude of the waveform is represented by the relative density 
of the ones and zeros within the bitstream;  for that reason the technique 
is also known as Pulse Density Modulation (PDM).  A full positive waveform 
results in all ones and, whereas a full negative waveform results in all 
zeros.  Silence would yield a string of alternating ones and zeros 
(Alternating ones and zeros reflect the "neutral" density of silence.)  Sony 
claims a resulting dynamic range of 120db for DSD.

In the simplest terms, DSD compares the energy of a waveform at each sample 
point to the accumulated energy of previous sample periods.  If the energy 
is higher, a value of 1 is recorded; if the energy is lower, a value of 0 is 
recorded.  Because the bitstream represents the change (delta) in energy 
compared to the sum (sigma) of the previous energies, another name for this 
technique is Sigma-Delta Modulation (SDM).
</quote>

The article goes on to explain how DSD uses a sampling rate of 2.8224 MHz 
resulting in a frequency response that extends to 100kHz.  The higher sample 
rate can also improve stereo/surround imaging.

<quote a="EM: A Better Mousetrap" p="96">
One more justification for DSD is that the steep filters required for PCM to 
conform to the Nyquist theorem can themselves degrade the sound due to their 
sharp cutoff.  In fact, it has become common for PCM analog-to-digital (A/D) 
converters to start with a 1-bit conversion stage and then apply a digital 
decimation filter to convert the bitstream to PCM format.  DSD essentially 
bypasses the decimation filter and stores the 1-bit data directly.
</quote>

DSD/PDM appears to be a superiour technique for recording and playing audio 
material.  Granted, this technology may never catch on because of all the 
hardware and software changes that would be required to mirror what a 
typical PCM based DAW currently does.  But, if DSD/PDM does catch on, and 
DAWs start being produced, how will this effect current audio DSP 
techniques?  How would JACK's current design be effected by a 1-bit stream 
format?

The article mentions a program called Pyramix (Windows) which features DSD 
support.  However, for Pyramix to do EQ, dynamics, reverb processing, and to 
display waveforms and vu levels, it converts DSD to a "high quality" PCM 
format.

I suspect that if DSD/PDM has any hope of succeeding in the current DAW 
world, there will have to be some real-time lossless converter constructed 
to seemlessly bridge the formats.  But that of course begs the question of 
why use DSD in the first place if you just are forced to "decimate" it PCM 
to process it?

Sony is currently using DSD/PDM in their SACD format, and are positioning 
themselves to be the competing format to PCM based DVD-A.  The fact that 
something like this shows up in EM means that at some level the format is 
being given some credence and is being watched by others then just the 
audiophile recording crowd.

Is DSD/PDM just a flash in the pan, or does it really represent a new wave 
of thinking regarding storing/processing audio data?

Dustin



_________________________________________________________________
MSN 8 with e-mail virus protection service: 2 months FREE*  
http://join.msn.com/?page=features/virus