I came across this site which give some background to the technology
behind the audio CD. Quite a bit of effort goes into trying to make sure
you don't hear the result of even fairly severe CD damage.

Greg Menke wrote:

Peter Fairbrother writes:

Any bits flipping because of analog effects is an error reading the
medium. It happens, which is why there are ecc and checksums. Their
job is to detect and/or recover from the error- the same kind of thing
is done in more expensive computer memory to help recover from cosmic
rays flipping bits here and there, among other things.

However, how the audio nuts go from arbitrary bits flipping between
states to an organized effect like "better bass" instead of a simple
increase in random noise is very curious to say the least.

A puzzler, but I can think of a few possibilities. Unlike data, the error
correction for music is CDs is not absolute - if there is an error the
circuitry will try and play around it - and the method used to approximate
that might well involve smoothing the sound out over a certain interval of
time with interpolated values, causing better bass and worse treble
response; or if not actually improving the bass and/or overall quality,
causing a relatively better bass/treble quality ratio.

In the studio I believe it is also an occasional practice to add noise at
innaudible level to ameliorate distortion caused by the notes being exact
fractions of the sampling ratio, and some non-linearities in the overall
response, causing and a slight increase in both actual and perceived
quality.

Note that these are pure speculation, and probably wrong. Any effect, if
there is an actual effect, may well be caused by other processes. They are
just to show that it might be possible.



The errored blocks of data are then missing and interpolated, which
inevitably introduces noise- sort of like a fleck of dust on the LP
groove. Its not just bass frequencies that are magically affected, the
lost samples literally stop the speaker cone from reproducing the
waveform for the duration of the error. At that point you're presumably
relying on the ballistic performance of the cones to "smooth over" the
lost samples and damp the sharp transitions from signal to error to
signal, which doesn't sound like a good way to improve "quality ratios",
whatever they are. Its effectively a decrease in the signal/noise
ratio.

The ECC codes are essentially data checksums, computed such that some
amount of missing or corrupt data can be recovered given the surrounding
data and the ECC code itself. Once the algorithm's capacity is exceeded
(because of too many errored bits in the associated data block), then
all you'll get out is data corrupted to some unknown extent. So from a
reproduction standpoint, you get perfect data from the CD as the number
of mis-read bits coming off the CD read head increases up to the point
where the bit-error-rate exceeds the capacity of the ECC algorithm, then
corrupt data starts coming out. By "perfect", I mean the bits read off
the CD are the same as written. So, the CD format provides a limited
degree of error recovery built into the format of the audio samples- it
alway reads the same bit-stream until too many errors show up for the
ECC to deal with.

I look forward to your detailed proposition documenting how bit flips
that get past the ECC algorithms improve sound, with emphasis on how CD
audio formats make it possible. I'm also interested in how bit flips on
dvd's and digital audio tapes improve the reproduction quality but do
not corrupt things like software images or compressed data.

Gregm