The Fifth Element #5 Page 2

Next comes digital word length: In a multibit system such as 16-bit/44.1kHz, word length determines how many fine gradations of dynamics can be preserved. The consensus seems to be that a word length of 16 bits is not enough, but that diminishing returns set in above 22 bits. (A resolution of 24 bits is much more frequently advertised than achieved.)

Let us not forget dither: In order to optimize the performance of the least significant bit in a multibit system, low-level noise must be added. The last several years have seen the development of very sophisticated proprietary dithering systems, and some professional processors provide for selection from a menu of various dither shapes. (Adding noise to increase resolution is about as counterintuitive as things get.)

Finally, in a mirror image of the input stage, at the final digital-to-analog conversion, another brick-wall filter is needed to keep the sampling frequency from intermodulating with the program material. It is felt by many qualified engineers that ringing from the D/A brick-wall filter is the prime culprit responsible for certain less felicitous aspects of conventional CD sound.

As this recapitulation demonstrates, there are at least five aspects of conventional digital recording, as it has historically been practiced, that may impose limitations on the system's ability to record and play back music. Some of these limitations were functions of the state of technology when the system was launched, but other limitations stemmed from the marketing decision that the CD had to fit into the standard DIN dashboard cutout yet still hold Herbert von Katajan's performance of Beethoven's Ninth Symphony on one side (footnote 6). So what about all those musicians who went gaga over early digital? Were they dupes, or just unworthy persons?

I think neither. In many cases, it boiled down to what they were listening for. Pianists in particular were delighted to hear piano tone reproduced without spurious vibrato added by analog tape flutter, print-through, or LP speed variations. String players were in many cases more cautious in their endorsements—they usually like more vibrato. Over the last 20 years, PCM recording has matured into a very sophisticated discipline, while many of the shortcomings of the CD itself have been worked around.

Of the five potentially problematic aspects of CD PCM, it would appear that we are stuck with the initial frequency cutoff of the anti-aliasing filter. Upconverting the sampling frequency cannot add the harmonic information the anti-aliasing filter originally kept out, but upconversion can nonetheless reduce the sonic degradation attributed to the complementary brick-wall filter at the other end of the process. We can tack on more zeros to make the digital words longer, but that can't add low-level information. The original dither can't be removed, but the data can be redithered. So we're stuck with four out of five, but sampling-rate upconversion can move brick-wall D/A filter artifacts up an octave or two.

However, the advent of DSD's very fast 1-bit system opens up at least the possibility that the final brick-wall filter stage can be avoided entirely, even for old PCM master tapes. The DSD sampling rate is so fast (64x that of CD) that an analog filter can accomplish analog conversion, albeit at the cost of substantial ultrasonic switching noise.

England's dCS has developed a pair of sampling-rate and format converters—the 972 for professional use and the Purcell for audiophiles; see Jonathan Scull's 972 reviews in the January and February 2001 issues—that can convert a data file from PCM to DSD entirely within the digital domain. Because the final brick-wall filter comes into play only on PCM's conversion to analog, converting an older PCM recording to DSD by purely mathematical means rather than by playing back to analog and then using the DSD process from scratch will have (at least in theory) sonic benefits, despite the limitations of the original recording process.

As a test subject for my comparisons of a DSD transfer of an LPCM original, I of course chose Glenn Gould's indispensable digital re-recording of Bach's Goldberg Variations (recorded 1981, released 1982). This touchstone recording has been reissued as a single-layer non-compatible SACD.

To eliminate to the greatest extent possible variability caused by playback equipment, it was necessary to use a player with incontestably top-drawer CD playback as well as SACD capability. John Atkinson generously gave up the Accuphase SACD transport and converter he had been enjoying in his system, and arranged for them to be sent to me.

I judged my standby Plinius CD-LAD line stage and SA-50 amplifier as equal to almost any high-resolution task, especially after David Blair of Custom Power Cord Company came through with the loan of Top Gun power cords for nearly everything, and a Top Gun HCFi for the power amplifier.

On setting up the two-box Accuphase SACD system, though, I did the obligatory forehead slap over the Accuphase converter's variable line output. Sure enough, running the signal directly into the SA-50 power amp was preferable to using Plinius' line stage. Creative Cable Concepts' RMC 12.0 SB interconnects ($1350/m), which I previously had shelved as a bit too revealing on lesser sources, worked like a charm on the Accuphase converter. My single-ended digital interconnect (for comparison purposes) was DH Labs' excellent Silver Sonic, while the CAT 5 cable that comes with the Accuphase duo handled DSD data transfer, and usually the "Red Book" data as well.

Next time: Choosing a high-performance monitoring speaker, and the proof of the pudding.

Questions? Comments? Praise and honors?.

Footnote 6: Good thing the piece that had to fit on one CD was not Mahler's "Symphony of a Thousand"; we would have been stuck with 12-bit resolution or 11kHz bandwidth.