Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.1 6/24/83 v7 ucbtopaz-1.8; site ucbtopaz.CC.Berkeley.ARPA Path: utzoo!linus!philabs!cmcl2!seismo!uwvax!harvard!wjh12!genrad!decvax!ucbvax!ucbtopaz!newton2 From: newton2@ucbtopaz.CC.Berkeley.ARPA Newsgroups: net.audio Subject: Digital companding and dynamic range (followup to Shaun) Message-ID: <569@ucbtopaz.CC.Berkeley.ARPA> Date: Fri, 12-Oct-84 02:31:43 EDT Article-I.D.: ucbtopaz.569 Posted: Fri Oct 12 02:31:43 1984 Date-Received: Sun, 14-Oct-84 05:41:36 EDT Organization: Univ. of Calif., Berkeley CA USA Lines: 39 Newsgroups: net.audio Subject: Re: Sheffield CDs and why they sound bad References: <460@watdcsu.UUCP>, <46@vice.UUCP> I agree with just about everything in Shaun's posting on dynamic range, but I'm a little uncertain about what one gains from using an explicit control word (the equivalent of a pilot-tone compander). The distortions of dbx-style companders are due to two sources: 1. The inherent distortion of the (analog) gain-control elements (the 20-cent VCA's) 2. The envelope-following distortion caused by the control voltage modulating the channel gain; this distortion is cancelled if the decoder tracks the encoder *exactly*. Aside from shoddy component tolerances, mistracking is inevitable in an analog system because the level-measuring circuits don't measure the same signal at the same time- it's been analoged somewhat out of shape via a nonlinear, non-flat and time-dispersive recording medium (ignore the matching of the supposedly "rms" level detectors). In a digital system, a feedback-compressor/feedforward-expander topology would be inherently *perfect* in the level-measuring/matching department, since both ends of the system would "measure" (Really just transcribe) the same digitized level data. So distortions due to non-complementarity would arise only from the step-size errors of the A/D/A converters. Again, I don't see what putting the control information on a separate track, as it were, gains you, since everything on that track is implicit in the digital data together with knowledge of the companding algorithm. You've got to compute the data for the control track from the data, so why not do it in the reproducer? If you want non-standard or variable algorithms, you could load them at the beginning of the selection (or periodically) but I don't think you need anything like 30% control-track overhead. The multiband observation is certainly a valid one, however, if you don't want to reintroduce all the breath, swishing and sucking artifacts of the pre-Dolby (and post-dbx :-)) era.