The Speed of Sound: How Safe is High-Speed CD-Audio Recording?
Robert A. Starrett
May 2000 |
Many of us yearn for the days when the pace of life was slower, when there were no firm deadlines in most everyday affairs and one could complete certain tasks at one's leisure, spend the rest of the time relaxing, and perhaps carefully correspond by letters sure to arrive within several weeks once mailed. Today gives us little of that; everything seems to move at Internet packet speed, and all those packets pile up in an unfathomable number of messages, files, programs, and spam. Remember when a letter delivered by post was an exciting event? You hoped it was filled with good news of friends and family; sometimes, of course, it contained bad news. But regardless of how long it took to get there, it came as news to you, since that was how information traveled in those days.
In CD recording, there is news aplenty, and it's all good. With today's 12X recorders, you can record a full audio disc in about six minutes. Compare that to the 78 minutes that it takes to record the same disc at 1X, and you can see why recorder manufacturers are scrambling to keep up in the "X" race. But bad news may loom as well, because some people question whether advanced recording speeds might have an adverse effect on the underlying data, the longevity of the disc, and its playability across multiple brands of CD-ROM and DVD-ROM drives and consumer audio and DVD players.
Today, our audio discs must play on more than an audio player. CD-ROM drives, CD recorders, DVD-ROM drives, CD-Audio players and changers, portable audio players, car audio players and changers, and DVD-ROM video devices all must be able to play back the data without a hitch, or else we consider the disc to be "bad."
Whether a disc sounds good, its longevity, and its playback compatibility over a wide range of CD-capable devices is largely dependent on error rates. All discs have errors on them. Luckily, most of these errors are corrected in playback. Physical error rates like Block Error Rate (BLER) can measure the quality of a disc. The final quality assurance in an audio disc is, of course, the listener's own ear.
As drive speeds rise--12X being the current state of the art--some wonder whether these faster recording speeds can accurately duplicate the source material without introducing artifacts that will spoil the listening experience. On the one hand, you can say that bits are bits and if they match, then there should be no difference in sound quality. On the other hand, some might argue that higher speeds might mean higher error rates overall, and the higher the error rate of a disc, the more likely it is that playback will reflect the errors, or that the longevity or interoperability of the CD will be compromised.
Do higher recording speeds result in more disc errors? They might, all things considered. Remember that disc error rates are dependent on the physical condition of the disc, and that physical condition is based on many factors--one of the most important of them being pit geometry. This is an area of study and even fascination for the so-inclined among us, who either get paid handsomely to dwell in such nanominutia, have too much time on their hands, or suffer from certain mental defects that make the subject interesting to them. Some of us even feel inclined to talk to our friends and family about it and alienate them all, perhaps resulting in involuntary commitment to a mental institution.
But the fact remains: if you're interested in audio recording, and you want to know your sound is safe, you'd be well-advised to acquire at least some basic sense of what to expect from the discs you record, and what it means when a good disc goes bad. You'll also have a clearer understanding of whether your disc-burning disappointments likely stem from faulty source discs or are endemic to the recording products or elements of the process, up to and including the speed at which you record sound.
DARKNESS AUDIBLE: HOW ERRORS ARE DETECTED AND CORRECTED
To understand the mystical side of things like pit-length deviation and pit slope, you must be committed--either to doing it or in the other sense. The data on CDs is physically composed of pits and lands, microscopic in size. Pits are bumps and lands are areas between the bumps. The laser in a CD playback device--whether an audio player, ROM drive, recorder, or DVD player--shines light on the disc as it spins. The diffusion of the reflected light from the disc is what determines whether a particular pit/land transition is interpreted as a one or a zero.
Whether the transition is interpreted as a one or a zero is dependent on the length of the pits. Pits are to be a certain length, varying from a 3T pit to an 11T pit. So, we have 3T, 4T, 5T, 6T, 7T, 8T, 9T, 10T, and 11T pits. Contrary to many people's understanding, pits and lands are not interpreted as ones and zeros. It is the transition between a pit and a land that produces a bit for transmission to the ultimate output of the drive.
The interpretation of the transition is dependent on the length of the pit. If a 3T pit is misinterpreted as a 4T pit-- because the laser has formed it a little too long--there is, of course, a physical error on the disc. On a CD-Audio disc, this error alone is not serious, because a player can mask the error in several ways. On a CD-ROM, the error cannot be masked, but must be corrected, if possible. This is because an error in computer data can crash a program, while an error in audio data is not as critical. This is why CD-ROMs have additional error detection and correction codes on them in addition to the CIRC, or Cross Interleaved Reed Solomon Code, that corrects errors on audio discs.
Before I start throwing "Es" at you, let's take a look to see what the E errors mean. Let's also have a definition of some of the other physical errors that can occur on a CD. And let's remember that for the purposes of this article, we have only looked at the number of E errors, Burst errors, and BLER overall. Whether and how high-speed recording will produce the other enumerated physical errors, and their effect on quality and playability, will have to be reserved for a future effort.
THE SWINGING Es
CDs use, at their base level of error correction, CIRC, or Cross Interleaved Reed Solomon Code. CIRC applies two levels of error correction, known as C1 and C2. The errors E11, E21, E31, E12, E22, and E32 derive from these two levels of error correction. CIRC employs two principles to detect and correct errors: redundancy and interleaving. CIRC uses about 25 percent data redundancy. The data is laid out in a way that will allow errors to be corrected, even though the data is not 100% redundant; it uses a parity-checking algorithm to reproduce data that is unreadable.
In addition, the data on a disc is interleaved. Interleaving means that the data is distributed over a relatively large physical-disc area. The data bytes are interleaved during recording and de-interleaved during playback. One data block (frame) of 24 data bytes is distributed over 109 blocks. To destroy one byte, you would have to destroy these other bytes. With scratches, dust, fingerprints, and even holes in the disc, there is usually enough data left to reconstruct any that has been damaged or caused to become unreadable. The C1 level of CIRC is meant to correct small, random errors. The C2 level corrects larger errors and burst errors.
THE E ERRORS, LEVEL 1
E11, E21, and E31 errors occur at the first stage of error correction. An E11 error means that one bad symbol (byte) was corrected at the C1 stage. An E21 error means two bad bytes, and E31 signifies three bad bytes were detected. E31 errors are uncorrectable at the first stage, and must be passed on to the second stage of correction. The first number is always the number of errors, and the second number is always the decoder level.
THE E ERRORS, LEVEL 2
In the second stage, C2, E12 means one bad symbol (byte) at decoder level C2; E22 means two bad symbols; and E32, three or more bad symbols. For CD-ROM, any E32 errors are unacceptable. First, an uncorrectable error in computer data can have a devastating effect. Second, the equipment used by mastering facilities will abort when an uncorrectable error on input media occurs, and the process must be restarted, which can result in additional costs and delays in your replication run. Some, but not all, CD-ROM drives use error correction and detection to such an extent that even an E32 error can be recovered, but there is no guarantee that even the best CD-ROM drive will be able to recover every E32 error. E32s on a pressed audio or ROM disc, or a CD-R, indicate that the user is dangerously close to losing stored data with the next scratch or fingerprint.
BLER stands for Block Error Rate. It indicates the number of data blocks that have any bad symbols (bytes) at the C1 error-correction stage. The reason that BLER is a good indicator of overall disc quality is that it essentially indicates the number of all errors, since errors that are uncorrectable at C1 get passed to C2. The Red Book specifies a maximum BLER of 220 per second, averaged over 10 seconds. Top-quality discs have an average BLER of below 10. A peak of 100 bad-data blocks per second is acceptable for CD-ROM, but an average BLER of 50 per second over the entire disc is a good cut-off point to ensure data integrity. CD-Audio discs can be acceptable with high BLER, but high BLER indicates that their longevity may be limited, and that there may be some problems in playing reliably on all types of readers.
Dropout indicates that the signal coming off the disc is below 30 percent of its nominal value. Dropout results from physical defects in the disc. There is no standard value for dropout.
When the laser beam is focused on a CD track, you can measure the difference between the light returned from the disc on the left and right sides of the track. When the laser beam is centered exactly over the track, the difference between the light reflected from the right side of the disc and the light reflected from the left side of the disc is zero. When the beam is off-center, push-pull tracking then becomes positive or negative. Push-pull is an indication of pit geometry and affects the overall gain of the tracking servo, which in turn affects the drive's ability to track the pits.
Eccentricity is the relationship of the geometric center of a track to the geometric center of the disc spindle hole. The CD-ROM standard is ±50 microns.
Reflectivity must be measured in relation to a known reference disc. Reflectivity is the variations of reflection from the reflective layer at different parts of the disc, and is especially significant for CD-R because the absorption of the light in the dye layer can cause a weak signal.
Birefringence (or double refraction) occurs when there are impurities in the polycarbonate of a disc, or when the polycarbonate is stressed when molded. The light is then refracted in more than one direction, and signal response to the optical pickup is degraded.
Jitter is a measure of errors in timing. When reading the transition from a pit to a land, there is an ideal timing window. The deviation between the window and the transition is important to the reading process. Jitter, as specified by the Orange Book, should be less that 35 nanoseconds. Since jitter can be caused by the player as well as the disc (i.e., the player may incorrectly perceive the transition), it must be measured in relation to a known reference disc.
Burst errors are consecutive bad blocks on a disc.
Table 1 shows error rates for a bad, out-of-spec audio disc.
THE BLER DISC PROJECT: SPEED KILLS, BUT WHO'S DYING?
To test the probability of error rates at different recording speeds, a sizeable quantity of audio discs from five vendors--Memorex, Imation, TDK, Sony, and Mitsui--were recorded at 1X to 12X speed and analyzed using testing equipment from CD Associates and Clover Systems. All results reported here were generated by the CD Associates CDA 1000. Additionally, discs were tested on a Clover QA101D as a control, and the results between the two testers were consistent.
Discs were burned on three recorders--Plextor 8X, 12X devices, and an HHB home recorder--at all speeds supported by the recorders and media. Discs were tested for error rates when recorded from a common-source disc. They were also tested against standard criteria for acceptable playback when burned at 1X, 2X, 4X, 6X, 8X, and 12X.
You can, of course, consciously make a bad audio disc. Take a cheap or damaged blank disc and record it at high speed. You can get a disc with really high error rates. Table 1 shows error rates from a bad audio disc, and the accompanying graphical representation shows how such a disc compares to standard disc-quality criteria. Note that BLER rates and E11 errors are significantly higher than the acceptable range.
Table 2 compares error rates for a "Clean" audio disc versus the same disc with multiple small scratches (MSS).
Table 2 shows, in Row 1, an excellent quality audio disc (denoted "Clean"). Row 2 represents the effects of applying multiple small scratches to the disc, which leaves it damaged in several respects but still meeting the minimum requirements. Row 3 shows the maximum error rates for keeping a disc within spec; note that E12 and E22 rates are beyond the acceptable level, while BLER is raised markedly but still not officially substandard.
Table 3 shows how varying levels of physical damage affect error rates of a high-quality disc.
The disc analyzed in Table 3 is also an excellent quality audio disc. Notice that there is no change between Row 1 and Row 2, which shows the effect of one fingerprint on the disc. Not too bad. Row 3 shows the effect of multiple fingerprints on the disc. Row 4 shows the effect of a big scratch. The scratch has pushed the disc out of spec.
When an audio CD player encounters an error, it uses either blanking or interpolation to "mask" the error. Remember that in audio playback, the affected bits are 1/75th of a second of sound. When blanking, the player simply does not play that 1/75th of a second of sound. Using interpolation, the player looks at what data came immediately before the error occurred and what data comes immediately after, and plays something that is close to both the previous sound and the following sound, thus masking the error.
The best overall measure of physical disc quality is the Block Error Rate, or BLER. BLER combines a number of disc parameters and gives us an overall feel for how well the disc has been recorded. Here is an interesting note: the BLER on pressed CDs that you buy at the record store is almost always significantly higher than that on properly recorded CD-Rs. Likewise, those AOL discs that you get every week also have significantly higher error rates than the CD-ROMs that you record at home, although they are still well within the Red Book specification.
Table 4 shows results of error-rate testing on five randomly selected pressed audio discs.
Table 4 presents a sampling of five randomly selected pressed audio CDs from my collection, and shows results from assessing BLER, E, and Burst rates for each of the discs. Notice the wide variety of BLER and E11 rates from disc to disc, all of which play more or less reliably with differences undetectable to the casual listener.
The discs charted in Tables 5 through 8 were recorded on a SCSI Plextor PleXWriter 8/2/20 attached to an AdvanSys 740UW PCI card, and a Plextor 12x4x32 using the same card. The test machine used with both recorders was a 450mHz AMD KII with 128MB RAM. The discs were all copied from the same CD-Audio source disc in Disc-at-Once mode with Prassi's Primo CD Pro 8.
Table 5 shows results of error-rate testing on Memorex media recorded at various speeds.
Table 5 shows results using Memorex media. Note that error rates at every level actually go down as recording speed increases, and the 8X recording actually yielded the highest-quality results of any of the speeds tested. While results dipped slightly for the 12X test, error rates remain extremely low and the discs eminently playable.
Table 6 shows results of error-rate testing on Imation media recorded at various speeds.
Table 6 shows results from recording with Imation media. Errors below the E11 level were almost nonexistent here, and BLER and E11 errors, both extremely low. Error rates increased slightly as the speed increased, but when your maximum BLER to stay in spec is 220 and your worst disc is 0.9, that six minutes you save seems well worth the sacrifice.
Table 7 shows results of error-rate testing on TDK media recorded at various speeds.
With TDK in Table 7, more of the same. Error rates are slightly higher across the board, but still way on the low side. Where there is a speed-to-speed to differential, note that the highest BLER and E11 rates were encountered at 4X, and results essentially leveled out the rest of the way, up to and including 12X.
Table 8 shows results of error-rate testing on Sony media recorded at various speeds.
Results encountered with the Sony media, shown in Table 8, again showed little difference in error rates as recording speeds increased. Again, 4X made the worst showing, but by quite a small margin. While the BLER spiked a bit at 12X, ask me, "Is it safe?" and I'll reply, "So safe you wouldn't believe it."
Table 9 shows results of error-rate testing on Mitsui media recorded at various speeds.
Mitsui has been as bold as any company in claiming that its media consistently works better at the high speeds for which the latest discs have been designed, and though the testing results with the Silver discs submitted were uniformly excellent, the 8X and 12X numbers did indeed equal the lower-speed performance specs in beyond-the-human-ear acuity as seen in Table 9.
As the tables show, all discs recorded at all speeds in this test showed extremely low error rates. There were no significant differences among the discs recorded at different speeds. It is useful to note that the difference between a 0.6 BLER, for instance, and a 1.2, while technically a doubling, is statistically insignificant and audibly non-existent, since the maximum allowable BLER is 220.
CATCH A WAV
Table 10 compares error rates when recording WAV files from a hard drive at 2X and 8X.
All the testing reported in the tables above was done with on-the-fly, disc-to-disc recording. To see if the uniformity of speed versus speed held true when recording from WAV files on a hard drive, Sony discs were burned from a common set of WAV files at 2X and 8X. Results from the WAV test to the disc at 2X and 8X shows that accelerating your recording speed makes little impact on disc quality, whether you record from a CD or from a hard drive, as Table 10 shows. Additionally, software makes no difference in the error rates. It can cause logical errors, but not physical ones.
YOU CAN'T GO HOME AGAIN
Table 11 shows error rates on four different disc brands when recorded at 1X on an HHB pro audio recorder.
Another claim heard from the audio community during its early dalliance with CD-Recordable technology is that audio recorders are the only way to go, whether because their 1X and 2X speed limitations keep things from getting too crazy or because some quality in their construction makes them more sympathetic to the audio task. To test this theory, discs from each brand in the test group were recorded at old, reliable 1X on an HHB Professional Recorder (which does not have the SCMS-only restriction of its "consumer" audio recorder counterparts). The discs were then tested for BLER, Burst, and six levels of E errors as seen in Table 11.
Finally, we see that recording at 1X from a CD source to CD-R on a professional audio recorder has no significant effect on the quality of the disc, whether measured purely through error rates or by the human ear.
So, the bottom line is that, when the world slowed down and we received our snailmail letter, delivered at 1X, it contained only good news and no bad news. But more importantly, by the time we got it, the news wasn't news at all--the 8X and 12X speedsters had beaten the pony express by a country mile and the message was the same: if you want to make your audio discs with the screamingest, fastest, baddest recorder on the block, go ahead. Let the timid confine themselves to recording at a snail's pace with the hope of improved audio quality. You know better.
Companies Mentioned in This Article
CD Associates, Inc.
15-A Marconi, Irvine, CA 92618; 949/588-3800; Fax 949/588-3805; firstname.lastname@example.org; http://www.cdassociates.com
31642 S. Pacific Coast Highway, Suite 101
Laguna Beach, CA 92677; 714/499-9566; Fax 714/499-4844; email@example.com; http://www.cloversystems.com
1410 Centinela Avenue, Los Angeles, CA; 310/319-1111; Fax 310/319-1311; firstname.lastname@example.org; http://www.hhb.co.uk
1 Imation Place, Oakdale, MN 55128-3414; 888/466-3456, 612/704-4000; Fax 800/537-4675; email@example.com; http://www.imation.com
Memtek Products, Inc.
10100 Pioneer Boulevard, Suite 110, Santa Fe Springs, CA 90670-3737; 800/636-8382, 562/906-2800; Fax 562/906-2877; http://www.memorex.com
Mitsui Advanced Media, Inc.
2500 Westchester Avenue, Suite 110, Purchase, NY 10577; 800/682-2377, 914/253-0777; Fax 914/253-8623; firstname.lastname@example.org; http://www.mitsuicdr.com
4255 Burton Drive, Santa Clara, CA 95054; 888/675-3986, 408/980-1838; Fax 408/986-1010; email@example.com; http://www.plextor.com
Sony Electronics, Inc.
3300 Zanker Road, San Jose, CA 95134; 408/955-5462; Fax 408/955-6822; http://www.sony.com
TDK Electronics Corporation
12 Harbor Park Drive, Port Washington, NY 11050; 800/835-8273, 516/625-0100; Fax 516/625-2940; http://www.tdk.com
Robert A. Starrett (firstname.lastname@example.org) is a contributing editor for EMedia, co-columnist for THE CD WRITER, and an independent consultant based in Denver, Colorado. He is the co-author of CD-ROM Professional's CD-Recordable Handbook.
Comments? Email us at email@example.com.