One of the usual responses to questions about quality of CD recording is "It's all ones and zeroes; quality doesn't matter". Some of this is discussed on a page here about media. The message there about indeterminacy of reading is the same here, but now with different implications.
When a CD is read, light from a read laser passes through the plastic, bounces from the metal reflecting layer, and is returned to the detector. On a pressed disc, pits and lands in the metal cause the light to be reflected well or poorly. On an erasable, the metal alloy's crystal structure makes the spot shiny or dull. On a write-once disc, the dye in the plastic causes the (infrared) illumination to be transmitted more or less well. In all of these, the distinction is not between zero and one, perfect reflection and perfect scattering, perfect transmission and perfect blocking. It is between less and more grayness. Somewhere in the continuum of grayness, the detector has a threshold: above that, call it a one; below, call it a zero.
Most of the time, the one-ness or zero-ness of the signal is correct. When it is uncertain or incorrect, Error Correction Code (ECC) is called on to fix it. With ECC, some of the available bits of storage are used to record information about blocks of bits so that if there are few enough errors in the block, they can be corrected. More error correction is provided on a data disc than on an audio disc since a glitch (a highly techncial term for an uncorrectable error) in audio will either be suppressed by the player or little noticed by the listener. But a glitch on a data disc may mean a program that doesn't work at all. (Video is comparable in this respect to audio so that for a VCD, the video file [DAT] is similarly given less correction code than a data file.)
Clearly, as the fuzziness of data reading increases, more and more correction is needed and the chance of uncorrectable error increases. If error correction can handle four mistakes in 2048 bytes, then if fuzziness gets up to an average of one in 2K, there will be a fair number of such sectors - there are over 300,000 of 'em on a full CD-R - with more than four, hence too many to correct. That data disc will not read correctly or reliably. "Reliably" because fuzziness varies with time of day, phase of the moon and supply of apples to the ancient gods - it varies unpredictably as the disc is read. When an uncorrectable error is first encountered, the drive will reread the sector in the hope that the next time is more propitious and valid data can be read. Of course, rereading means stopping the usual forward progress of reading, backing up the optical assembly and synchronizing again; in short, it really slows things down. When a disc has enough errors, reading slows substantially, sometimes to *less* than 1x.
Two significant things occur here. First, you can guesstimate the quality of a recording by the speed with which your drive reads it. Note that in a different drive, the threshhold will be different, the fuzziness will vary and the read will go at a different speed. Thus, you can't really say that it's a bad disc; all you can say honestly is that you cannot read it well in a given drive.
The second implication of rereading is that if matters get really bad, the drive may keep trying 'forever'. That is, the system may hang while the drive tries in vain to get valid data. It is a property of the basic -86 chip design and its interrupts that only one basic I/O command will be recognized at a time. While waiting for a read to be acknowledged, the chip will not 'see' a keyboard request. If the read is repeated in the hardware until a good signal is found and that signal never arrives, the keyboard is locked out. Only power-down or the equivalent reset will get through to stop the fruitless effort to read a disc that's bad enough.
The bottom line on all of this is that errors happen. They are not extraordinary, diabolical or traumatic; in reasonable numbers, they are expected and corrected, but that correction comes at a price. The more fuzziness in the data, the greater the chance that an error will escape correction, that the disc will read slowly or even that it will hang up. Thus, our objective is not only to write with no uncorrectable errors - the only ones which will be caught automatically - but to write with as few correctable ones as practical. If a signal level of 0 is a perfect zero and 1 is a perfect one, we want to have mostly .1, .2 and .8, .9 values detected and very few in the .4-.6 region. As noted in the page on media, those detection levels vary with the reader, but if you write consistently strong signals, the disc is likely to work well on all drives which are reasonably compatible with your medium. (That is not to say that an old Hitachi drive will read phthalocyanate media well; it won't regardless of how well it is written. But you can do nothing about that except change medium.)
If you're writing audio, fuzziness in writing - high error rate - degrades performance but usually does not cause failure. In playback, some drives compensate (see the page on Losses) by reducing expectations. All will provide at the analogue output error concealment which can cost high-end response to avoid noises which would be objectionable. That sort of correction is not applied on digital output, so for easy listening you may want to use analogue signal through your sound card even though many modern applications let you play back the more accurate digital signal. Most significantly, if you are extracting digitally, you will find spikes corresponding to uncorrected error - very annoying clicks due to a misread and uncorrected bit. A happy result of all this is that if you have a batch of media which you cannot record with low error rates, you may still be able to get rid of them for low-fidelity audio discs which are intended for listening, not for extracting. Since error rates tend to go up on the outer portions of a disc, they are especially useful when used for short, low-fi audio. So even a poor purchase can eventually be disposed of, though only with care.
Whether writing audio or data, you want to minimize the risk of poor reading. To do that, you want to minimize the number of correctable errors in what you write. You can do nothing about the range of readers that will be used other than choose media which are recognized as being of high quality, but among those choices you can select those which have the fewest errors when written by your drive of choice at the speed you use. Unfortunately, the quality of writing varies with the speed at which bits are written and in modern drives that speed varies over the disc. Every writer calibrates the power of the write laser at the beginning of a recording and good ones recalibrate during the write, but there is an optimum speed at which the fuzziness of bits is least and writing is best and no recalibration will keep the system at that optimum throughout.
Clearly, then, we want to find a way to test our writes to determine how good they are - how many correctable errors they have. To get an idea of how difficult that is to do thoroughly, drop by Media Sciences at http://www.mscience.com/ and read up on it. The tests that they perform are thorough and demanding - and far beyond the hardware and software available to the home user. Even a lower level of testing, Block Level Error Rate or BLER, requires special drives and equipment. On the other hand, simple pass/fail tests with your mastering software provide no information on correctable errors. In between, there are tools which will help. Since a crude test of the number of re-reads provides insight, you can judge qualitatively by the speed of extraction (audio) or copying (data). Far better is the measurement provided by programs such as Exact Audio Copy and CDSpeed. My own choice is for the Arrowkey products CD/DVD Diagnostic and Inspector which are explicit and, for Inspector, graphic. (Links are on the page of URLs, of course.)
Finally, I note that how well you need to write is up to you and to your application. If I'm spending $1000 to have a disc pressed, I'm willing to measure my master before I send it off, even though a thorough test will take ten minutes or so. When I restock my media, I buy enough to make testing worthwhile and check out a few samples at different speeds in my drives. Those which give zero errors - correctable and not - are used for archiving. Any with even low error rates are used for less critical applications. But in no case would I trust my data to a medium I had not calibrated for my purposes.
You may want to do something similar.
E-mail me at cdrecording@mrichter.com
Return to Mike's home page