Recording to CD: More than Meets the Ear
Page 1, 2, 3, 4, 5

 

Error handling: Prevention, Correction, and Concealment
As I've said earlier, the pits and lands in the CD track are incredibly tiny -- only 0.5 microns wide, which is approximately the wavelength of the pickup laser itself! Because they're so small and are passing by the reading laser so quickly, it's easy for the pickup system to misread a bit or two every once in a while, even under the best of conditions. Add to the environment a bunch of fingerprints, a few dust-bunnies, and the inevitable scratch here and there, and you've obliterated whole chunks of data! It's easy to see that in the real world -- where audio engineers routinely eat pizza over consoles, leave coffee rings on track sheets, and leave CDs gathering moss on their hopelessly cluttered desks -- an optical medium like compact disc would be useless without a way to gracefully recover from errors caused by conditions like this. Audio CDs use a combination of clever techniques to handle errors -- from little ones to really big ones.

Parity checking
Parity bits are added to each frame, both prior to and after the data is interleaved (more on interleaving in a minute). By checking the parity bits, and comparing them to what the laser saw (or, at least thought it saw), a CD player can tell if any bits were misread, which ones they were, and what the correct value should have been.

Parity checking is good for catching small errors, but doesn't help if there is a surface defect which covers enough of an area to wipe out the parity bits themselves. Large errors which cause whole bunches of data to go missing are called burst errors. That's where interleaving takes over!

Interleaving
Imagine if you will a written page of, say, 40 lines. Let's also say that someone took a bottle of White-Out and, dabbing letter by letter, obliterated 4 lines in the middle of the page. You now have a printed page of words, only with a huge chunk missing.

As a reader, how likely are you to be able to reconstruct that missing information in your brain? The fact is that your chances aren't very good. Now, maybe that missing chunk of information isn't very important in the context of what's being said. On the other hand, it could be crucial! You might be able to gather the gist of the story without those lines; and then again, you could be hopelessly lost. Whatever the case, the fact remains that there's no way to reconstruct those missing lines word for word.

Now take the same situation -- but first scramble the letters on the page before dotting White-Out on the middle four lines, and then de-scramble them afterward back to readable form. Note that the same amount of information was obliterated. However, what's happened now is that the missing letters, instead of being one contiguous chunk, are now scattered all over the page. What you've got is the equivalent of a puzzle, like on the game show Wheel of Fortune! Using contextual information and your knowledge of the language, your chances of being able to fill in the missing information are greatly improved. (And who knows... you might even win a trip to Cancun! Um, okay, probably not.)

This is a simplified explanation of how interleaving works on a compact disc. The information doesn't get written to the data track in the order it appears, but instead the recorder first scrambles and shifts the original data around so that it doesn't appear in sequential order on the CD. It does this using an interleaving algorithm that the player understands, so that the player can read the data one disc revolution at a time and "de-interleave" the data on the fly during playback. If there is a smudge of mozzarella on the CD (hey, you know who you are) that renders a contiguous series of bits unreadable, those missing bits will be "scattered" after the de-interleaving process. The player can then use algorithms to help it figure out what the missing bits should have been.

Interleaving allows even larger errors to be corrected than with parity alone. In fact, using parity checking and interleaving together, a good player can correct a burst error of up to 4,000 bits or so. That's approximately one inch of the spiral data track.

Quite often, a burst error (or a series of them) occurs which is even too large for parity checking or interleaving to take care of. Interpolation to the rescue!

Interpolation
Interpolation is exactly what it sounds like. When there's a lot of data gone bye-bye, the player will figure out an "average" based on some good bits that come before and after the bad bits, and then substitute the average where the missing data should have appeared. Because musical waveforms are pretty much continuous over the short run, the player can usually do a good job of "connecting the dots" as it were. Depending on how accurate the interpolation is, the result can be anything from a totally transparent continuation in the audio signal to a slightly audible glitch.

Interpolation isn't truly an error-correction technique, but is more accurately an error "concealment" technique. Interpolation can conceal errors of about 13,500 contiguous bits, which is about 3 milliseconds or so of audio.

Muting
CIRC encoding is an incredibly effective process, reducing the error rate on an audio CD to about 1 uncorrectable bit in 1,000,000,000. Add interpolation to that, and what you have is a very robust medium indeed!

Despite all of this, however, there can be burst errors too large for even CIRC and interpolation to fix. When all else fails and a player encounters an error it can't handle, it will simply mute the output for a fraction of a second. To avoid a pop or glitch, the player actually does an extremely fast "fade-out-fade-in" to cover things up.

It's interesting to note that, despite the complex error-correction methods used by audio CDs, error-correction on CD-ROMs (data or "Yellow Book" CDs) is even more complex! This is because while an unrecoverable error on a audio CD might at worst result in an annoying skip during playback, a similar error on a CD-ROM might corrupt a crucial data file, which could easily render an entire software package useless. Of course, the more robust error-correction technology required by CD-ROM takes a lot more room to encode as well. So, if you've ever wondered why a recordable CD can hold 74 minutes of digital audio (which is over 783 megabytes in size) but only 650 megabytes of CD-ROM data, now you know why.

Go to Page 5; Back to TOP