jj@alice.UUCP (06/04/85)
> From allegra!ulysses!mhuxr!mhuxn!charm!prk Wed Dec 31 19:00:00 1969 > > ... > 1. Good digital systems have noise levels which are much > smaller than the least significant bit. That's why you > can, in principle, copy over and over without ANY degradation. > My bet is that the signals in digital recording satisfy this > condition by a large margin. This point, if that's what it is, is orthogonal to the question of degredation. See my previous posting. > 2. Computer memories reserve a significant fraction of their > space for error-correction bits. They suffer from having > bits shot out at random by ionizing radiation, chiefly due to > cosmic rays and radiactive impurities in IC packages. With > the right error-correcting code, however, the mean time between > uncorrectable error can be made much longer than the obsolescence > time of the machine. Scientific American published a nice little > article about this within the last two years. Precisely. > 3. About digital recordings: Even though CD-player blurbs talk about > "error-correction", I'll bet there is none - there's no time You're very, very wrong. There is significant error correction power in the simplest of the CD players. In the most fancy ones, they use ALL of the 80% redundancy in the stored data to fix things. > for it. The real issue is, if there is an incorrect bit now and then, > could you hear it ? NO! A single glitch in a digital record YES! A single MSB inverted shows up as a gargantuan, earsplitting !!!CLICK!!! The fact that you never hear any CLICK's shows just how well the error correction works. (In fact, if things are that bad, it zeros thee sample, rather than CLICK!ing at you.) In fact, the MSB's are a bit more protected for exactly that reason. > leads to white noise of very low amplitude, as I have verified by depends on what bit you hit. > toying with artificial data sets simulating my experimental Um, I don't want to all you a liar, so send me mail and let's see what's wrong with the process. It's a standard test that I've carried out on every form of digital coder compressor, or direct audio that I've worked on, and PCM is the most sensitive of the bunch. > data. You can't hear it. Think of this another way. Suppose > Beethoven is playing along, and then one bit gets out of place. > This causes a click of duration 20 microsec. How sensitive > would you be to that? Not very sensitive. These glitches Quite sensitive. Sorry to be so directly contradictory, but your article exactly contradicts my experience. > would have to be quite frequent to be annoying. They're not - > ... > Granted that one hit of an LSB wouldn't make any difference, still, one hit of an MSB causes quite a ruckus. If you meant to say that the effect after error correction was essentially nill, that's a different story. -- TEDDY BEARS HAVE LIMITED PATIENCE! THEY DO EVENTUALLY GET HUNGRY! "Let us remember my cat, Geoffrey, ..." (ihnp4/allegra)!alice!jj
hsu@cvl.UUCP (Dave Hsu) (06/06/85)
> > From allegra!ulysses!mhuxr!mhuxn!charm!prk Wed Dec 31 19:00:00 1969 > > > > ... > > 2. Computer memories reserve a significant fraction of their > > space for error-correction bits. They suffer from having > > bits shot out at random by ionizing radiation, chiefly due to > > cosmic rays and radiactive impurities in IC packages. With > > the right error-correcting code, however, the mean time between > > uncorrectable error can be made much longer than the obsolescence > > time of the machine. Scientific American published a nice little > > article about this within the last two years. > > Precisely. > > > 3. About digital recordings: Even though CD-player blurbs talk about > > "error-correction", I'll bet there is none - there's no time > > You're very, very wrong. There is significant error correction power > in the simplest of the CD players. In the most fancy ones, they > use ALL of the 80% redundancy in the stored data to fix things. > > > for it. The real issue is, if there is an incorrect bit now and then, > > could you hear it ? NO! A single glitch in a digital record > > YES! A single MSB inverted shows up as a gargantuan, earsplitting > !!!CLICK!!! The fact that you never hear any CLICK's shows just > how well the error correction works. (In fact, if things are that > bad, it zeros thee sample, rather than CLICK!ing at you.) In fact, > the MSB's are a bit more protected for exactly that reason. > ... > > data. You can't hear it. Think of this another way. Suppose > > Beethoven is playing along, and then one bit gets out of place. > > This causes a click of duration 20 microsec. How sensitive > > would you be to that? Not very sensitive. These glitches > Quite sensitive. Sorry to be so directly contradictory, > but your article exactly contradicts my experience. > > would have to be quite frequent to be annoying. They're not - > > ... > > > Granted that one hit of an LSB wouldn't make any difference, > still, one hit of an MSB causes quite a ruckus. If you meant > to say that the effect after error correction was essentially > nill, that's a different story. > -- > > (ihnp4/allegra)!alice!jj The bottom line is...that's not the reason ECC was ABSOLUTELY necessary. When 'soft fails' (i.e. cosmic radiation etc. -caused errors) first came under recent scrutiny, it was estimated that the average home computer at the time suffered perhaps 2 soft-fails per week. At this rate, the error become almost insignificant, but ECC would be desirable. The obvious reason for ECC is that famous culprit...the long blob of scum. Whether from a defect or from surface contamination, you will get much more than 20 usec of noise. In fact, since track number, running time, and other information are evidently encoded on each track, ECC becomes absolutely necessary or your player might go berserk trying to find where it was. You don't believe the effects of minor errors? There's one eluding my correction scheme on a Chopin Etudes disk; it always bounces back to the same track. Thank heavens (and technology) for the CD. Goodbye to the Rice Krispies guys hiding out in my records. -dave
jj@alice.UUCP (06/10/85)
Sorry, Dave, that might be the "obvious" reason to you, but it's not the reason for ECC's at all. It's the reason for TIME DIVERSITY (i.e. the cross interleave, but not the ECC). In any case, it's necessary to have both ECC and time diversity. The "blob" graphically demonstrates the reasons for diversity, and the fact that the average player does ECC twice a second on a GOOD disk demonstrates the reason for ECC. To put it otherwise, there would be a full scale CLICK every 8 seconds on the average if there wasn't. That click, alone, would render the disc completely impossible. The blobs would, too, of course. There are many papers in ICASSP and IEEE-Com Transactions that talk about errors in PCM codes of various sorts. Read one and you'll see what I mean. -- TEDDY BEARS HAVE LIMITED PATIENCE! THEY DO EVENTUALLY GET HUNGRY! "Let us remember my cat, Geoffrey, ..." (ihnp4/allegra)!alice!jj