ECC, manufactured data, and blue LED's
Charlie Hand
The French guy with the blue LED's is a hard case, to be sure, but a
couple of things he said are not very unlike comments which are made
from time to time on RAHE, and to clarify things I thought I'd go over
the process of error-correction in CD players. My own ECC design
experience has been for data streams other than CD, but I have a CD
technical reference here to guide me, and hopefully if I make any
mistakes someone will be good enough to point them out.
There are two catagories of error handling in a CD player, error
correction and error concealment.
Error correction is magic. Well, not really, but it takes quite a bit
of frowning and head-scratching to follow the boolean math which
allows errors to be corrected. The process is not intuitive, and
unless one is an extraordinarily crafty person, one cannot make very
many correct assumptions about how it works without actually studying
how it works.
Error correction does just what it's name says - it corrects errors.
Or not, but it always does one or the other - either it corrects
errors or it does not. Error correction doesn't guess what the correct
data might be, nor does it smooth over errors or distort them into
something which sounds like a non-error. It either sends out the exact
same data that was recorded on the CD, or else it sends out an error
flag indicating that there is an error which it cannot correct.
Error correction, especially in CD's (where there are actually two
error correction algorithms going on at the same time), is a
relatively BIG digital process. There are dozens of flip-flops and
hundreds of logic gates, all madly changing state from one clock to
the next. The process is essentially the same when there are errors in
the input stream as when there are no errors in the input stream. In
either case, all the hundreds of logic gates are madly changing state
from one clock to the next. There is no special error-correcting block
of circuitry which is suddenly fired up when an error comes along and
shut down there are no errors. It doesn't have to "work harder" when
errors are present. It's all or nothing. CD players have error
correction circuits, because the standard says so, and those error
correction circuits run all the time. Whether the input data is
error-free, or has correctable errors, the error-correction circuit
runs essentialy the same, and it's output is bit-for-bit perfect in
either case.
When the error-correction circuits receive non-correctable errors,
then they behave slightly differently, but only very slightly. Inside,
they function essentially the same as if there were no errors or
correctable errors (gates and flip-flops still maddly changing state
every clock), but the output is slightly different: it has incorrect
data on it, and the incorrect data are tagged.
These non-correctable errors go to the error concealment circuit which
makes a guess, and a pretty cotton-picking good one, as to what the
value of that bad 44/1000 of a second should have been.
Different people quote different numbers, but these non-correctable
errors are rare in the extreme. For a reasonably clean CD, and unless
your CD player is *really* screwed up, you can expect from zero to ten
non-correctable errors over the length of a CD. That's maybe ten
44/1000's of a second out of 160 million 44/1000's of a second on the
CD - perhaps one 44/1000's of a second every six minutes. Rare in the
extreme!
Now, one 44/1000 of a second interpolated out of every 6 minutes does
not manifest itself in the imaging, or the soundstage. It doesn't show
up in the liveness or the spirit or the authority of the sound. As a
scientist, I never use superlatives, but I'll make an exception in
this case. The non-correctable errors in a CD player are ABSOLUTELY
inaudible.
To summarize there is an error-correction circuit which runs all the
time, errors or no errors, and which produces aproximately
99.99999% of the samples bit-for-bit exactly as they were recorded on
the CD. The other ten or so (out of 160 million or so samples on a CD)
are interpolated by the error-concealment circuit, which also runs all
the time, errors or no errors, which gates it's result (which it
calculates even when it doesn't need it) into the data stream for
approximately one 44/1000 of a second every six minutes.
IN A CD SYSTEM, DATA ERRORS ARE NOT AN ISSUE!!!
Please pardon me for yelling.
-Charlie
Charlie Hand
In article <59bmme$g...@tolstoy.lerc.nasa.gov>,
Charlie Hand <cha...@hand-family.org> wrote:
>
>These non-correctable errors go to the error concealment circuit which
>makes a guess, and a pretty cotton-picking good one, as to what the
>value of that bad 44/1000 of a second should have been.
Ah, shucks, did I keep saying 44/1000 of a second? Make that 1/44,000
of a second.
-Charlie
--
--
/Newsgroups: .*,.*,.*,/h:j
Scot D. Pencil, M.D., Ph.D.
Charlie Hand wrote:
> how that bad 44/1000 of a second should have been.
I am sure you meant: bad 1/44,000 of a second... a much smaller time
interval. Probably everyone who thought about it for a moment
corrected this for you. I wonder why the RAH-E editors did not?
-Scot.
[ Moderator Note: I approved the original and it looked fishy, but I
figured everyone would catch the drift. -- rgd ]
brad sanders
cha...@hand-family.org (Charlie Hand) sez:
>.... That's maybe ten
>44/1000's of a second out of 160 million 44/1000's of a second on the
>CD - perhaps one 44/1000's of a second every six minutes. Rare in the
>extreme!
Ironically, Charlie, the "correction" may still be inaccurate. This is
because the data are also in "frames" - and these "frames" run closer
to 7Kz than 44KHz. Many CD recovery circuits can't tell the outside
world *exactly* where, within a frame, the bad sample actually is (The
Philips chips come to mind here).
So, depending upon how the transport handles it - and how the DAC
handles error flags - you may get one sample interpolated... or you
may get a frame of samples (a coupla cycles at 20KHz) "stretched," or
chopped away.
In less than optimum circumstances,
creativity becomes all the more important.
------------------------------->
>http://radioactive.home.ml.org<-----------------------
jj, curmudgeon and all-around grouch
In article <59bmme$g...@tolstoy.lerc.nasa.gov> cha...@hand-family.org writes:
>There are two catagories of error handling in a CD player, error
>correction and error concealment.
Yep.
>Error correction is magic. Well, not really, but it takes quite a bit
>of frowning and head-scratching to follow the boolean math which
>allows errors to be corrected.
>
>Error correction does just what it's name says - it corrects errors.
Yep. Think of it this way. You take the data, and you add more
information that depends ONLY on the data. This is called
"redundancy". This redundant information means that if some of the
information gets lost or broken, you can still reconstruct the exact
signal, and that means EXACT signal.
>It either sends out the exact
>same data that was recorded on the CD, or else it sends out an error
>flag indicating that there is an error which it cannot correct.
Yes, and it can also tell with very high accuracy when the correction
fails, in case somebody is going to ask.
>every clock), but the output is slightly different: it has incorrect
>data on it, and the incorrect data are tagged.
Usually, MUCH incorrect data. Basically, error correction of most but
not all sorts can remove up to n% errors. Any errors more than that
get reflected by at LEAST ER-n percent if not more. Where ER is the
error rate percentage.
This means that using error correction, you usually go from perfect to
perfectly horrid, with no space in the middle.
>These non-correctable errors go to the error concealment circuit which
>makes a guess, and a pretty cotton-picking good one, as to what the
>value of that bad 44/1000 of a second should have been.
And this isn't specified as well as it might be. Most of my CD's never
activate this. I dunno how good it is, because any real interpolation
has to be rather bad, by nature.
--
Copyright alice!jj 1996, all rights reserved, except transmission by USENET
and like facilities granted. This notice must be included. Any use by a
provider charging in any way for the IP represented in and by this article
and any inclusion in print or other media are specifically prohibited.