marcel@ecn.uucp (Marcel Bernards) (10/09/89)
Hello DSP Guru's, I have a question related to digital recorded music, CD's and a possible lack of information in the reconstructed signal. A few weeks agoo, there was an interview in a Duch paper about a studio engineer who claimed that somethin _Very_ essential was missing in digital recorded and mastered music. He said that he had listened for a long period to analogue and digital recorded music and strongly got the feeling that something was missing in the sound. He felt that old recordings, even those from 20 years agoo sounded more realistic and live than digital recordings, especially classical solo piano music. Therefor he designed a 'black box' that reconstructs the 'missing part' of the digital information. As an electrical engineer and guitar player I'm very interested what the missing part of the digital signal should be. Maybe it's a Phase relation problem or some sampling problem ?? Is there someone who can explain what is _really_ missing ? Or is this guy just a charlatan who wants to make a lot of money ?? Greetings, -- Marcel Bernards, UNIX & Net sysadm Netherlands Energy Research Foundation ECN P.O. Box 1, 1755 ZG Petten, PHONE: 09 312246 4579 EARN/BITNET:ESU0130@HPEENR51 EMAIL: marcel%ecn@{nluug.nl,uunet.uu.net},marcel@ecn.uucp ..!hp4nl!ecn!marcel SCREAMNet : AAAAAARGHH!HUH?? : Disclaimer: "The AntiChrist is the Computer !"
jbuck@epimass.EPI.COM (Joe Buck) (10/15/89)
In article <1989Oct9.082931.9020@ecn.uucp> marcel%ecn@nluug.nl (Marcel Bernards) writes: >I have a question related to digital recorded music, CD's and a possible >lack of information in the reconstructed signal. > >A few weeks agoo, there was an interview in a Duch paper about a studio >engineer who claimed that somethin _Very_ essential was missing >in digital recorded and mastered music. By the Nyquist-Shannon sampling theorem, a band-limited signal can be reconstructed exactly from its samples if it is sampled at more than twice the highest frequency. To do this in practice, you would need a perfectly band-limited signal on the input end, and an ideal low-pass filter on the opposite end (or, equivalently, an ideal low-pass filter on both ends). We also lose accuracy because we're representing the amplitude of a signal by a 16-bit number, but that's still a huge dynamic range compared to what you get with an LP; the major design problem with CDs is in the anti-alias filtering. First, the recording end. If you screw up here, no black box can do much to save you. Some of the first CDs were really shoddy jobs; this really shouldn't be much of an issue any more. Since once the information is in digital form, perfect copies can be made, the best CDs are those in which the original recording is done digitally, with well designed equipment. On the playback end, we have a D/A converter and a low-pass filter. If the D/A converter put out Dirac delta functions and the low-pass filter had a (sin t) / t response that went backward in time, we'd get perfect reconstruction of the signal. We don't. A typical D/A puts out a staircase-shaped function, which introduces a bowing effect at high frequencies, and then we require an analog low pass filter that passes 20 KHz without attenuation, but cuts off 22.05KHz (half the CD sampling frequency) almost entirely. It's very tough to design such a filter. Good CD players incorporate DSP tricks to solve both these problems. To solve the steep low-pass filter problem, oversampling is used. Oversampling means that zero samples are inserted between the samples, we do digital low-pass filtering (cheap and easy), and then we need a much gentler analog low pass filter. So CD player manufacturers want to charge you extra for oversampling, even though it saves them money to do it! (if they didn't oversample they'd have to spend more money on analog filtering). To handle the "bowing" distortion caused by the staircase shape of the D/A response, the signal can be pre-distorted digitally to compensate. This is easier to do on an oversampled version of the signal. > Therefore he designed a 'black box' that >reconstructs the 'missing part' of the digital information. I wouldn't be surprised if his "black box" contained just what I described above: oversampling and "predistortion" to compensate for the effects of the D/A converter. But I doubt he is doing anything unique. >...Or is this guy just a charlatan who wants to make a lot of money ?? Could be. I've seen a few "audiophiles" who basically claim that the sampling theorem is invalid or that digital recording is "mechanical" by its very nature and cannot convey emotion or other stuff like that. -- -- Joe Buck, just visiting/consulting at Entropic -- write me at: jbuck@janus.berkeley.edu ...!{uunet,ucbvax}!janus.berkeley.edu!jbuck