home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!uwm.edu!linac!att!att!allegra!alice!jj
- From: jj@alice.att.com (jj, curmudgeon and all-around grouch)
- Newsgroups: rec.audio
- Subject: Re: MD and DCC encoding-request for info
- Message-ID: <24516@alice.att.com>
- Date: 30 Dec 92 17:44:26 GMT
- Article-I.D.: alice.24516
- References: <shetline-271292211904@128.89.19.80> <DAVE.92Dec30234952@pipi.iis.u-tokyo.ac.jp>
- Reply-To: jj@alice.UUCP (jj, curmudgeon and all-around grouch)
- Organization: NJ State Home for Bewildered Terminals
- Lines: 54
-
- In article <DAVE.92Dec30234952@pipi.iis.u-tokyo.ac.jp> dave@pipi.iis.u-tokyo.ac.jp (David Wuertele) writes:
- >In article <shetline-271292211904@128.89.19.80> shetline@bbn.com (Kerry Shetline) writes:
- >> If they [MD or DCC] only handle PCM data, would there be a generational
- >> degradation when copying MDs/DCCs (forgetting about serial copy mgt for the
- >> moment)? It seems to me quite likely that there could be -- calculation
- >> round-offs, inconsistent "framing" of the PCM data stream...
- >
- >Yes, there will be if there is any type of vector quantization going on in
- >the compression (block transform + quant schemes are a subset of all
- >vector quant schemes), because the PCM data stream has no framing
- >information to keept the vectors consistent. There is also a possibility
- >that pre- and post- filtering is conducted, which will degrade even more.
- >
- >Dave
-
- Oh, sheeesh.
-
- Both coders (ATRAC and MUSICAM/PASC/ISO Layer I and II) are lossy
- perceptual coders.
-
- They both work on the same general principle, using this sort of block diagram:
-
- --------> Filter Bank --------> Rate Control ----> Bitstream generation
- | ^
- | |
- | |
- -----> Perceptual Model -----------
-
-
- In MUSICAM the filter bank is a 32 band polyphase filter bank
- by Dehery et. al after Crochiere and Rabiner "Multirate Digital
- Signal Processing", the Perceptual Model is a Zwicker-based
- model (See ISO-MPEG-1 Draft Audio Standard), the rate control
- is based on 3 groups of 384 samples in time/frequency, and the
- bitstream uses PCM and some minimal radix encoding to do
- transmission/storage.
-
- In ATRAC the filter bank is an MDCT (See Princen and Bradley's
- ICASSP paper 1987), the perceptual model isn't published, the
- rate control is some kind of block companding, and the bitstream
- unpublished. The MDCT is switched in length, it's not clear what
- choices are supported presently.
-
- A good place to read about this stuff is "Advances in Speech Signal
- Processing", Furui and Sondhi, Chapter 4, by Brandenburg and Johnston.
- Marcel Dekker, NY 1992. For newer stuff see the latest ICASSP,
- look for papers by Johnston and Fereirra, Davidson et al, Singh,
- and others. The Johnston paper was submitted late, it's on the
- last four pages of the audio book.
- --
- Extremism *Copyright alice!jj 1992, all rights reserved, except transmission
- in the *by USENET and like facilities granted. Said permission is
- defense of *granted only for complete copies that include this notice.
- liberty is no vice. *Use on pay-for-read services specifically disallowed.
-