Listening Through the 16-bit Window

Home | Audio Magazine | Stereo Review magazine | Good Sound | Troubleshooting

Digital engineers have recently held out the promise of improved resolution from the 16-bit CD medium. Many strategies have been suggested, some compatible with the existing, widely established format, while others require new specifications and new player designs. However, with one new format—Sony’s MiniDisc— technology advances have been used to reduce the resolution available, to get a long-term payback in the form of smaller and cheaper music data storage. MD is widely felt to sound inferior to high-quality CD playback.

A more recent proposal concerns the high-density digital video disc (DVD) formats, developed by Sony/Philips (responsible for the CD format itself) and Toshiba. The objective was to get a whole feature film onto a single CD-sized disc. With CD offering just 0.65 gigabytes of storage capacity, the potential offered by a disc with up to 10-Gb available is immense. I hope the two rival groups join forces and agree on an industry-wide standard for both video and high- quality audio.

Professor Malcolm Omar Hawksford, the Director of the Centre for Audio Research and Engineering Department at the UK’s University of Essex, has proposed that the audio community fight for the next step in digital playback quality? The increase in storage density would allow this relatively straightforward technology—no exotic blue—light lasers required here!—to provide an audio disc offering a doubled 96kHz sampling rate as well as a digital word length of 24 bits. (The latter is the present studio standard.) Advances in data coding and handling would allow this new disc to carry audio data in a variety of forms: multi-channel sound; CD standard 44.1kHz, 16-bit; and high resolution. Different decoders of varying cost and ability would be able to choose both how to read the data and to what performance level.

The 24-bit, 96kHz-sampled disc would be the “gold standard” for at least the next 20 years. The sonic difference between such a high-resolution disc and CD would be comparable to the difference between a 30ips, half-track master tape and a Dolby-B analog cassette. This would be a true audiophile CD well worth fighting for.

THE HDCD ALTERNATIVE

A high-resolution audio disc based on DVD technology is a far more radical proposal than the Pacific Microsonics HDCD system? HDCD encrypts extra information about the signal in a buried data channel. It therefore requires a unique encoder—and, for full compatibility, a unique decoder—to play recordings made to that standard. While HDCD has sonic merit, especially for properly decoded recordings, it represents an unwanted complication for the CD format, in my view.

There is also the question of compatibility. Despite what is claimed, my experience has been that HDCD recordings replayed without decoding do not always sound “better” than a normal recording. Pacific Microsonics claims that the HDCD code artifacts are inaudible and the encode process improves replay quality, decoded or not. On the several recordings I have auditioned undecoded, I have heard low- level artifacts, shifts in reverberant die—away, changes in timbre, and some compression of both the dynamic range and of stereo image perspectives—all of which must be regarded as distortions of a kind (4). While many will feel such additions to the signal to be minimal, it is a matter of concern for the vast majority of us—those without HDCD decoders—that these coded recordings could sound “altered” compared with conventional linear coding.

NOISE SHAPERS

There is another way to radically improve the sound quality of the 16-bit CD as we know it without any need to change our players and decoders or to fit compatible filter/decoder chips. This readily applicable technique is dither, applied with noise-shaping. Much has been said about this process— by shifting quantizing noise up to the inaudible 20kHz region as it reduces the output word length, it preserves as much as possible of the original’s resolution in the midrange— but its full implications have not really been spelled out.


Apogee’s UV-22 uses dither close to the Nyquist Frequency (22.05kHz in the CD system) to preserve resolution. Meridian’s 618 noise shaper. Both input and output word lengths can be set independently, and a choice of several noise-shaping algorithms is offered.

Such processes — Sony’s Super Bit Mapping algorithm, DG’s (Deutsche Grammophon) 4D process, the Apogee UV-22, and Meridian 618 — are applied to a high-resolution digital recording at the final stage of its transfer to a 16-bit (213 master. While there are differences between the various techniques, they do have a common objective. Taking the human hearing response into consideration, they can and do increase the replay resolution of the 16-bit CD to 18 or even more bits, at least in the midrange and low treble. In terms of the preservation of low level detail, ambience, and timbre, the gains are greater still.

It’s not widely appreciated that until recently the vast majority of the available CD repertoire whatever the production process and whatever the high number of bits used in the studio, is brutally truncated—chopped off—to raw 16-bit code to prepare the CD master tape or CD-R. Fundamentally, truncation is the sound of CD as we know it. I believe we have all been guilty of unwittingly confusing the sound of ”digital” with the sound of 16-bit truncation (whether this is due to poor AID performance or poor code handling in the editing/mastering process).


Fig. 1. A noise shaper shifts noise from the lower frequencies, where it will be audible, to the top octave, where it becomes inaudible (despite its rise in level) due to the human hearing system’s reduced sensitivity in this region. With a high-quality DAC, the psychoacoustic result is an increase in signal resolution beyond that possible from a non-noise-shaped 16-bit digital system. (After Meridian.)

LISTENING TESTS

I recently carried out lengthy listening tests to determine the “sound” of 16-bit digital audio. While analog playback gives a point of comparison, it suffers from its own characteristic imperfections. What should be the baseline for the comparisons?

Recording engineer Tony Faulkner provided me with some top-quality 24-bit recordings. (He had used the new DCS Delta/Sigma ADC, which is monotonic to 20 bits and continues to output useful, music-related data down to the 24th bit.) In replaying these recordings from his Sony PCM-9000 magneto-optical master disc recorder, the data were truncated to the 16-bit resolution to simulate typical CD replay and fed via an S/PDIF data link to a digital decoder with a clean 16-bit performance. (Fed -90dB dithered tones, the replay distortion was better than 120dB down relative to peak level.)

To obtain 16-bit data with higher resolution, I used the Meridian 618 processor, whose own signal transparency had been carefully established. Cross-checks between the 24-bit master and an appropriate noise-shaped dither mode on the 618 showed that, indeed, much of the full quality of the high-resolution recording was available from the 16-bit version.

A REVELATION!

The listening comparisons were revelatory. I hadn’t expected truncation to 16 bits to affect all major aspects of subjective fidelity—tonality, image quality, and clarity on both low and high volume settings. But it did. The corrupted sound of truncated 16-bit digital audio was clearly audible, and no, it wasn’t the “digital” part we found wanting, it was the “16- bit” part.

What we are familiar with as discrete 16—bit digital was audibly imperfect. Compared with the noise-shaped version, it sounded gray, grainy, hardened, thinned, mechanical, dead, and dry. And this was only in the loud passages—the bits where, even in a 16-bit system, it is commonly supposed that there is so much resolution available that reproduction will be near-perfect. Low-level detail was also masked, reverberation tails truncated, and low—level timbres were shifted and roughened.

And there was more! The effect of truncating the die-away of ambience and musical notes damaged both stereo focus and stage width and depth. The size and scale of the stereo image and the whole soundstage, the ambient space, was reduced. And some of the sense of flow and rhythmic structure of the music was impaired.

On one recording of a piano duet, the master tape featured a generous stage width and a palpable 5’ of separation between the performers. In “16-bit truncated” mode, easily half the Roslyn Hill Chapel acoustic was lost while the timbrally distorted images of the performers jumped closer together, lending a near mono quality to this recording.

Increasing the number of encoded bits from 16 to 18, 20, and 24 provided a significant gain in sound quality. At the 24-bit word length, the master recordings no longer sounded “digital while combining the promised virtues of digital— clean, tight, unphasey bass; pure dynamic peaks; rock-steady focus; zero wow; and quiet noise floors—with all the musical purity and beauty of fine analog.

These recordings firmly drew listeners into the musical experience. They were excellent in terms of image size, focus, space, and depth. Orchestral strings had natural timbres, while brass instruments had the right combination of richness and edge rasp. Flutes no longer sounded excessively breathy, and the hall acoustic could be heard clearly, successive cascades of notes playing into the reverberant field with their pitches preserved. A sense of air and clarity pervaded the acoustic.

To quantify the improvements heard between “normal 16-bit” and dither-encoded 18- or +20-bit resolution, you could pay thousands of dollars for such a sonic difference between two CD replay systems.

THE 78-RPM MONO EXPERIENCE

There is an extraordinary twist to this story concerning another engineer’s experience when transferring some old 78rpm discs to CD. It was thought that the resolution of a 16-bit A/D converter would be more than sufficient for a single-channel medium with ostensibly a 40dB dynamic range and a 100Hz—8kHz frequency range—and that’s on a good day! In theory, even a dithered 8-bit channel would be sufficient to carry all the data present on a 78!

Yet truncation from a nominal 20-bit word length to 18 bits on these 78 transfers was audible as a change in timbre. Thus, even with original analog recordings of apparently limited quality, proper use of rounding and noise-shaped dither is required if the maximum sound quality is to be achieved.

If a truncation at the 18th bit is audible for a 78rpm mono signal, then how much more important must it be for a modern stereo signal to be treated with due responsibility?

CRIMINAL ACTS

The recording industry needs to become aware of the value of higher-resolution recordings. When engineers say, as many still do, that digital operations such as level control, equalization, and truncation to 16 bits must be inaudible for domestic replay because any artifacts are better than 90dB down from peak level, believe me—they’re wrong. All kinds of digital manipulations are routinely carried out without regard for the introduction of quantization distortion due to truncation. With regard to high fidelity, the truncation of higher-resolution recordings to 16 or fewer bits is criminal.

This matter can be addressed simply. In order to preserve transparency and tonality, recording engineers must work with the highest practicable number of bits, and use dither for all digital signal processing. The reduction to 16-bit CD code must be held off until the very last stage in preparing a CD master. That final step has to be made with the application of appropriately noise-shaped dither to transfer the maximum of recorded information to CD with the widest possible resolution and dynamic range for consumer replay.

A MATTER OF EMPHASIS

Although the CD standard provides for complementary high—frequency emphasis and dc-emphasis, this has fallen out of favor. Very few CDs are encoded with HF emphasis. Yet when executed in the analog domain, following a CD player’s D/A conversion stage, de-emphasis can provide around 1 bit, 6dB, of reduction in replay noise and distortion. Many player’s and D/A processors implement de-emphasis in the digital domain; i.e., as part of the digital filter (over-sampler). This, however, can impair performance due to the additional computation necessary. Even without de-emphasis, sensible use of noise-shaped dither will push 16-bit replay to better than 18 bits , with subjective transparency sufficient to still hear the effect of bit changes at the 24- and 20-bit levels.

Notes:

4. In part, I believe the basis for people liking the sound of undecoded HDCD recordings is the gentle compression employed, which makes the sound “louder” in a rather benign manner. On the other hand, the high sampling rate and long word length offered by the ADC section of the proprietary I-IDCD encoder will also play a part in the claims of improved sound quality.

5. Provided the replay DAC has at least 18-bit resolution and dynamic range. The measurements in reviews will tell you which these are.

6. To enable readers to make such comparisons for themselves, tracks 13 and 14 on Stereophile’s Test CD 3 are the same music recording—Canadian pianist Robert Silverman performing a Chopin Waltz—reduced from 20-bit to 16-bit resolution using the Meridian 618, and truncated to 15-bit resolution, among other things. See the ad elsewhere in this issue for ordering information.

Also see: HDCD Explained

(Article courtesy STEREOPHILE, June 1995; by M. Colloms)

Top of page   Home


Updated: Saturday, 2009-06-13 3:35 PST