jj@alice.UUCP (06/02/85)
There are several differences between copying an analog master and a digital master. When you copy an analog master, you irrevocably add a certain noise power (depending on the recording medium) to the recording, regardless of what the signal used to look like. Basically, assuming random noise in each copy, you get a certain signal to noise ratio in the first copy and you loose 3dB for each doubling of the number of copies, assuming the same recording medium under the same conditions. In addition, any elements of frequency and/or phase domain distortion are added linearly (in dB and radians), hence the frequency shaping can become extreme very fast. When you copy a digital master, the signal to noise ratio determines the percentage of bits that's wrong. For an SNR of 20!! dB, the normal sort of modem/etc, with average bandwidth/bitrate, will run at about 10^-6 error.(Yes, I'm approximating here, but I'm in the right ballpark. Given a reed-solomon encoding of that, the actual chance of not catching that error (parity doesn't work for correction errors, but other encodings, such as the ones that CD's and PCM-F1's use, do, rather well) is essentially more than squared, (And I'm being VERY conservative in saying only swuaring!) i.e., giving a bit error rate in excess of 10^-12. That's for a really bad SNR on the recording medium. What's more, the system is regenerative, in that a second generation copy may have a bit error somewhere (if it's long enough), BUT!!! the SNR at the time of recording is again perfect, i.e. the noise from the previous recording pass is gone, and only the errors remain, thus (using our 20dB number), the SNR of the COPY is also 20dB, in the recording medium. In other words, the bit error rate of a copy is the same as the original. For those confused by the 20dB number, Imust note that the digital and analog (by digital, I mean the SNR of the recording medium, nad by analog the output signal from the whole process) are totally unrelated, untilthe SNR of the digital end drops to a low value such that significant bit errors start to occur. While I don't feel like regurgitating all the math here, the bit error rate is essentialy 50% until the SNR gets to a breakpoint, and then drops rapidly, very much like the outlier probability for 1 sigma, 2 sigma, 3 sigma, etc. <There's reason for the similarity, of course> So, there you have it, a short talk on SNR in analog and digital recordings. -- TEDDY BEARS HAVE LIMITED PATIENCE! THEY DO EVENTUALLY GET HUNGRY! "What's the use of bearing bracers, hats or spats or shoes with laces, or the things they buy in places down on Brompton Row? (ihnp4/allegra)!alice!jj