You may be aware of Super Bit Mapping (SBM), Sony’s CD mastering technique. The technique reportedly provides nearly 20-bit audio quality from the 16-bit Compact Disc.
After demonstrations of SBM at the Chicago Consumer Electronics Show and at Sony’s Tokyo recording studios summer 1992, I and everyone else who heard the comparisons agreed that SBM was a significant improvement over conventional transfer technology. It seems, however, that there may be more to the SBM story than first revealed—both in SBM’s technology and marketing.
At the most recent Audio Engineering Society convention in San Francisco (full re port to follow next month), a Sony engineer presented a technical paper on SBM. The presentation ignited a firestorm of controversy that continued in the hallway for 45 minutes after the paper’s conclusion. Many respected audio scientists questioned both the technique and how it has been represented to the audio press.
On a technical level, it appears that SBM isn’t as sophisticated as was first suggested. The promises of having additional information below the 16-bit level (not accounting for dither) and a smaller quantization step size aren’t quite accurate. Instead, SBM is merely a noise-shaping technique that pushes quantization noise away from the midband (where the ear is most sensitive) to the upper treble (where the ear is less sensitive). Period. Al though this is a valid scheme and an improvement over no noise-shaping—pushing the noise down below what would be the equivalent of an 18-bit system, where the ear is most sensitive, results in improved resolution —a similar technique has been an integral part of the Harmonia Mundi Acoustica encoder for two years.
Much of the criticism at the AES focused on SBM’s noise-shaping curve—a curve that reportedly mimics the human ear’s varying sensitivity to different frequencies. Professor Stanley Lipshitz—whose former student, Robert Heddle, now works for Sony and gave the presentation—sat next to me and offered his comments throughout the paper. Of Sony’s curve, he said, “I don’t know what creatures that curve was designed for, but it certainly wasn’t for human beings!” Dr. Lipshitz had prepared a graph comparing his own noise-shaping curve (developed over two years ago) with the SBM curve (fig.1). The Lipshitz curve shows a much lower noise floor between 2kHz and 6kHz, a region where the ear is most sensitive. According to K. Hazaina, a Sony engineer involved in SBM development, Sony considered the Lipshitz curve (and others) but chose their own curve after extensive listening tests. The Sony curve looks less good on paper, but reportedly sounds closer to the 20-bit digital source than the other curves evaluated.
Fig. 1: Comparison of Lipshitz and Heddle (Sony) noise-shaping curves. Note
that both trade off a large rise in audio-band noise above 14kHz for increased
resolution below. The Lipshitz curve offers the greater increase in HF noise,
but more closely mimics the threshold sensitivity of the human ear and gives
significantly greater resolution in the low treble where the ear is most sensitive.
(Linear frequency scale, 10dB/vertical div.)
Dr. Lipshitz also objected to the fact that SBM has garnered so much attention when the idea isn’t as revolutionary as press reports indicated. Lipshitz published work in this area over two years ago. His argument that the SBM curve is not optimum may be a case of differing opinions among researchers. However, his feeling that SBM has received undue media attention and hype is justified, in my view, considering that the technique has been known for some time.
Similarly, several other researchers approached me to express their dismay that SBM has gotten so much hype Further, they objected to the Sony-supplied plots printed in my first reports One graph shows a 1kHz sinewave reportedly with and without SBM. The SBM-processed waveform appears to have a much smaller quantization step size— an impossible feat when the signal ends up as 16-bit. It was publicly charged at the paper presentation that this constituted deliberate deception: the SBM waveform had been low—pass—filtered for the graph, while the non-SBM waveform had not, it was alleged Moreover, the noise-shaping curve supplied by Sony showed only the noise spectrum up to 10kHz, hiding SBM’s huge increase in noise level between 10kHz and 20kHz (see fig.1). Sony spokesperson Marc Finer went on the record to state that both 1kHz wave forms published in our report were subjected to identical low-pass filtering.
Robert Adams of Analog Devices suggested that SBM-processed CDs will sound worse on CD players with poor low-level linearity. He argued that, in a fade to silence, the silence is no longer a static digital code with SBM. Instead, many LSBs will be toggled, continually crossing the zero transition where linearity error is greatest, particularly if the DAC’s MSB trimmer is misadjusted or nonexistent. This condition will add high levels of white noise to the signal. It should be noted, however, that high-end CD player and D/A processors I’ve measured (including both 1-bit and RI2R ladder DACs) have shown excellent low-level linearity—usually less than <1dB error at —90dBFS and often less than <0.5dB at —100dBFS. Although Mr. Adams’s concerns are theoretically valid, it seems a backward step not to use SBM because some users with poor-quality converters may hear more noise, thus denying the sonic improvements of SBM to more critical listeners who own even moderately good converters.
[While at the dbx corporation, Bob Adams designed the first and, to date the only 128x-oversampling, 20-bit, analog/dig ital converter. UltraAnalog bought the rights to the design and supplies the module for high-end A/D conversion applications (the Manley and Wadia ADCs, for example). It has remained to this day the only 20-bit A.DC extant. Bob Adams’s original engineering prototype is the heart of the Chesky/Bob Katz converter heard on all Chesky releases. ]
Other objections to SBM were raised at the presentation. Some criticism, however, was clearly motivated by commercial interests: one vociferous critic has a competing CD mastering system of lower quality on the market.
Another facet of the story reported to me was that SBM has been in use for nearly two years on Sony Classical productions— without even a mention on the liner notes or to the press. Indeed, Richard Schneider’s “Industry Update” in last November’s Stereophile (p.4 suggested that SBM has been in use since January 1992—and used on all titles in Sony Classical’s Leonard Bernstein
Royal Edition series, Kathleen Battle’s and Wynton Marsalis’s Baroque Duets, and Emmanuel Ax’s Brahms Variations disc. It has been speculated that SBM was conceived as an in-house technique for improving Sony’s digital transfers, but was elevated to “break through” status and vigorously promoted when Sony realized that they had to differentiate the sound qualities of CD and MiniDisc for commercial reasons. It could be coincidence but SBM and MD have been promoted concurrently. According to Finer, several test CDs were made using SBM as long as two years ago, but were not commercially re leased in this country. He also investigated the Kathleen Battle and Emmanuel Ax master-tape genealogy and confirmed that SBM was not used on those titles.
Finally, it should be noted that, despite the controversy, and whatever Sony’s motivation, SBM does indeed improve CD sound quality. Moreover, it is the first public acknowledgement by one of the CD’s inventors that conventional 16-bit digital audio has room for improvement. How much SBM improves CD sound quality will be more apparent after we have firsthand experience of the process; Sony has invited us to transfer the analog master tape of Stereophile’s most recent recording project to digital using SBM. It will thus be possible to hear the original analog master, the full 20-bit dig ital transfer, and the SBM-processed 16-bit. We may even include samples on our next Test CD so that you may judge SBM for yourself After all, the proof is in the listening.
== ==
ALSO SEE:
Deutsche Grammophon 4D recordings
== == ==