EFM and CIRC

A compact disc is designed to be handled by people who eat while they do it. That single requirement is why roughly half its physical capacity is spent on not being wrong. The problem is not that errors happen; it is their SHAPE. A fingerprint two millimetres across destroys about seven thousand consecutive channel bits. A single-error-correcting code does nothing at all against that, and making the block bigger makes it worse, because a bigger block catches more of the burst. The answer, and it is one of the most elegant ideas in engineering, is to stop the errors being consecutive BY THE TIME THE DECODER SEES THEM. Write the data through a set of delay lines of staggered length so that symbols which were neighbours in the message end up scattered across the disc. A burst on the disc then arrives at the decoder as one damaged symbol in each of many codewords, and one damaged symbol is exactly what a modest code can fix. Cross-Interleaved Reed-Solomon Code — two codes over GF(256), (28,24) and (32,28), with the delay lines between them — corrects a burst of about four thousand channel bits, which is two and a half millimetres of track, completely and bit-exactly. Alongside it, Kees Immink's eight-to-fourteen modulation does the channel coding that blueprint 7 covered — with one extra job. An optical pickup detects a THRESHOLD on a reflected intensity, so a drifting DC level moves the threshold and corrupts every decision at once. EFM spends three merging bits per symbol, carrying no data whatsoever, keeping the running digital sum near zero. That is why it looks wasteful next to RLL(2,7) and is not.