The $2000 CD Made from Glass
Japanese CD manufacturer Memory-Tech is offering to replicate CDs made from glass rather than the conventional polycarbonate. The rationale? Glass CDs sound better.
Memory-Tech calls the discs “Crystal Disc” and so far, seven titles are available in Japan with more on the way. Deutsche Grammophon has released its 1963 recording of von Karajan conducting Beethoven’s Ninth symphony and has plans to release two other titles. EMI’s entry is a recording of the Japanese violinist Mariko Senjyu. JVC will soon offer three titles by Japanese classical artists. All the titles are mastered with JVC’s K2HD process.
Judging from the initial titles the discs seem intended primarily for the Japanese market. The pricing also seems aimed at the ultra-top-end Japanese consumer; as much as $2000 for a single disc. I suspect the reason that the discs are so expensive is that each one is individually cut on the laser mastering machine rather than being replicated from a stamper.
Memory-Tech sent me two discs containing identical data, one made traditionally from polycarbonate and the other from glass. The disc is a sampler of different musical selections, chosen to reveal sonic differences across a broad spectrum of music. The glass-substrate CD is heavier and stiffer than the plastic disc, but is otherwise similar.
But how can two discs containing the identical ones and zeros sound different?
I’ve long been fascinated by the idea that a CD’s optical properties can affect the sound in an analog-like manner even though the datastream remains unchanged. My interest in this subject began in the mid-to-late 1980s when I worked in a CD mastering lab. In addition to being part of a six-man team that designed and built CD (and Laserdisc) mastering machines, my job included correlating problems in replicated discs with anomalies on mastertapes and the mastering process. (I co-wrote, with Ray Keating, an Audio Engineering Society paper on this subject called “CD-V Signal Optimization.”) Incidentally, we had a process for making one-off CDs on a glass substrate, but it was used primarily to check the programming of a CD-ROM in the very early days of that format (1986).
A CD-replication client claimed that the discs we made sounded inferior to his mastertape. I performed a bit-for-bit comparison between the mastertape (3/4" U-Matic tape in the PCM-1630 format) and the replicated disc using a CD-ROM pre-mastering computer. Not surprisingly, the data were bit-for-bit identical. I was unable to verify the client’s claim that the disc sounded different from the master because I couldn’t listen to both sources through the same D/A converter (the Sony PCM-1630’s digital interface is a format called SDIF-2, which uses three BNC-terminated lines—left channel, right channel, and word clock).
My colleagues—all “bits is bits” engineers—dismissed the client’s claim as mere delusion in light of the bit-for-bit accuracy of the disc to the mastertape. In their view, we had done our job in delivering a CD with a bitstream that was identical to the master. What more could one ask for?
Nonetheless, I wanted to pursue this question, and cut another master from the same tape, but on a different mastering machine. The client reported that the disc made from this second master sounded significantly better than the first disc. With two CDs, I could now listen to them for myself decoded through the same D/A converter. The client was correct; the second CD sounded smoother, more spacious, and less “digital.” He didn’t describe the differences in those terms; to him, the first disc was simply missing his musical expression.
There were no manufacturing differences between the discs; neither had uncorrectable errors or other problems that are routinely checked during QC. My curiosity was piqued, so I had the jitter on both discs analyzed using a specialized piece of test equipment. To understand the concept of jitter in a CD, some background on how the CD works is necessary.
Digital data are stored on the CD in “pits” (indentations in the disc) and “lands” (the flat disc surface). The transition from pit-to-land or land-to-pit represents binary “one.” All other surfaces (pit bottom or land) represent binary “zero.” The pit and land structures don’t represent the data directly. Rather, an encoding scheme called “eight-to-fourteen modulation” (EFM) creates patterns of data in which successive binary “ones” are separated from each other by a minimum of two “zeros” and a maximum of ten “zeros.” This produces nine discrete pit and land lengths on the disc.
The playback laser beam is reflected from the disc to a photodetector that converts light to an electrical signal. The nine discrete pit and land lengths produce an amplitude-modulated signal at the photodetector composed of nine discrete sinewaves, which vary in frequency from 196kHz (corresponding to the longest pit or land length) to 720kHz (corresponding to the shortest pit and land lengths). The digital data are contained in the sinewaves’ zero-crossing transitions.
The jitter analyzer counts the exact frequency of each of the nine sinewaves and then graphically plots their frequency distribution. The distribution is Gaussian, with most of the pit and land lengths falling very close to the ideal. On the disc that sounded inferior, the distribution was extremely wide, with large variations in the pit and land lengths. On the better-sounding disc, the distribution was sharply defined and the curve was very narrow. In other words, the first disc had a greater amount of jitter encoded in the physical structures that represent the digital data. You could see this by looking at the signal from the photodetector; the so-called “eye pattern” was a little ragged on the first disc compared to that of the second disc. (There’s a reason the second disc had lower jitter; the mastering machine on which it was cut featured a completely different turntable design and more sophisticated rotational-servo control than those of the first mastering machine.)
Note that the pit- and land-length variations were not great enough to be interpreted incorrectly; a binary “one” was never mistaken for binary “zero.” The datastreams were identical after decoding.
I once pressed a CD optical engineer at Philips on this question and came away with the distinct impression that he understood the mechanism by which discs with identical datastreams sounded different, but wouldn’t publicly admit the phenomenon (for obvious reasons). The engineer gave me knowing smile and a wink, repeating the party line that CDs were incapable of analog-like variability in sound quality. If anyone knew the answer to this mystery it would be Philips; it contributed the optical aspects to the CD format (Sony developed the error correction, integrated circuit design, and hardware manufacturing processes, broadly speaking).
Getting back to glass-versus-polycarbonate CDs, it’s worth noting that polycarbonate can introduce optical distortions that affect the playback laser beam. Specifically, polycarbonate can cause a phenomenon called “birefringence”—a double refraction of the playback beam introduced by variations in the refractive index of the material through which the beam is passing. These variations in the refractive index are caused by localized stress on the polycarbonate introduced during injection-molding of the disc. That is, the liquid polycarbonate didn’t flow properly into the mold, creating areas that introduce birefringence. Obviously, a glass-substrate CD doesn’t suffer from this problem.
So how does Memory-Tech’s glass-substrate CD sound? I hate to rely on that old cliché in describing improved digital sound, but the glass CD sounded more “analog-like.” The glass CD was smoother, more spacious, more open, deeper, and had greater ease. By comparison, the polycarbonate CD was flatter and had less air between images; instrumental textures were less natural, sounding slightly synthetic by comparison. The polycarbonate CD by contrast overlaid timbres with a patina of glare. The difference was significant, but at 100 times the cost of a conventional CD, glass CDs will appeal to a very small minority of audiophiles.
This additional experience of hearing differences between CDs with identical datastreams makes it clear to me that the quality of the signal at the photodetector affects the disc’s sound. I don’t know how variations in the eye pattern find their way into the analog output signal—the photodetector’s output undergoes a huge amount of decoding, error correction, de-interleaving, and other processes to extract the raw PCM audio data that are converted to analog by the DAC. Nonetheless, there’s no question in my mind that a disc’s optical properties, which directly influence the eye pattern, introduce an analog-like variability in sound. The mystery remains.