The Analog Compact Disc Page 5
Fig.7 CD surface on which an apparently innocuous scratch has been made with a ball-point pen. Note the large amount of degraded data.
Fortunately, the CD format uses a very powerful error-detection and -correction system that nearly always results in bit-for-bit accuracy in the recovered data. In fact, the error-correction system is so effective that it can correct for up to 4000 consecutive missing bits. I'm not talking about error concealment, in which missing data are guessed at, but complete correction of those 4000 bits.
Errors can be corrected because data additional to the required data are already recorded on the CD. This so-called "redundant" data is called upon if the primary data are missing or corrupted. The most basic form of error-correction coding is simply storing the data twice; if the data are missing in the first location, you can recover the missing bits in the second location. This is an extremely crude example. In practice, error-correction schemes are enormously complex and vastly more efficient. The team of engineers and mathematicians at Sony who developed the CD's error-correction system reportedly spent ten years at the task.
A common misconception holds that data errors degrade sound quality. As shown later in this article, uncorrectable errors are rare events. However, even if you had a disc with hundreds of uncorrectable errors, those errors wouldn't affect such aspects of sound quality as treble smoothness, soundstage depth, or bass definition. Instead, you would hear a click or discontinuity at the point where the uncorrectable error occurred. The rest of the music would be unaffected. Moreover, there's absolutely no evidence that discs with lower error rates (corrected errors) sound any different from discs with high error rates. Whatever's causing differences in sound quality between CDs isn't data errors.
Nonetheless, knowing the number and severity of data errors gives an overall indication of the disc quality, and also shows if the disc has been abused.
The Cross Interleaved Reed-Solomon (CIRC) error-correction scheme used in the CD has two levels of error correction, identified by the names of the two decoding circuits: "C1" and "C2." The C1 decoder corrects for short, easily correctable errors; the C2 decoder handles longer "burst" errors that are harder to correct.
To understand how this two-step process works, you need to know about an error-correction technique called "interleaving." Interleaving mixes up the order of the data before recording, then puts them back in the correct order (de-interleaving) on playback. Interleaving converts long burst errors into many smaller errors, which are more easily corrected. In the CD system, de-interleaving occurs after the C1 decoder but before the C2 decoder. If an error is too long for the C1 decoder to correct, the C1 passes it to the C2 decoder, with a flag identifying the bad or missing data. The de-interleaving process between the C1 and C2 decoders makes that error much easier for the C2 to correct. Errors corrected by the C2 decoder are thus more severe than those corrected by the C1 decoder.
Errors are classified by how many bad or missing data packets (called "symbols") exist, and at which decoder they appear. We thus have a two-digit number that tells us the severity of the error. For example, an E21 error means that two bad symbols were corrected by the C1 decoder. The first number tells us the number of bad symbols corrected, the second number the decoder. An E11 error (one bad symbol at the C1 decoder) is very easy to correct; E11s happen all the time.
Conversely, an E32 error is uncorrectable; the data loss is so severe that the system cannot replace it. When this happens, the C2 decoder fills in the missing samples by interpolating (guessing at) what data may have been lost. Early CD players simply muted the audio output when uncorrectable errors were encountered. If a disc contains no E32 errors, the recovered data are bit-for-bit identical to the source data.
A general indicator of disc quality is called the "Block Error Rate," or BLER. BLER is the number of data blocks per second containing bad symbols at the C1 decoder input. The Red Book specifies a maximum BLER of 220/second (the CD data structure contains 7350 blocks/second).
The error type—BLER, E11, E22, for examples—gives us an indication of the error's cause. A high BLER indicates poor pit geometry: the optical system has a hard time reading the disc, and consequently produces lots of random bit errors. E22 errors (the worst fully correctable error) indicate localized damage to the disc, caused either by defects introduced by the manufacturing process, or mishandling of the finished CD.