Xref: utzoo sci.electronics:6955 rec.audio:13912 Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!mailrus!cornell!vax5!33581544 From: 33581544@vax5.CIT.CORNELL.EDU Newsgroups: sci.electronics,rec.audio Subject: Truth about "18-bit" digital audio products Summary: What the consumers don't know really won't hurt them Message-ID: <19037@vax5.CIT.CORNELL.EDU> Date: 15 Jul 89 19:44:13 GMT Sender: news@vax5.CIT.CORNELL.EDU Reply-To: max@london.ee.cornell.edu (Max Hauser) Followup-To: rec.audio Distribution: na Organization: Cornell, Ithaca NY Lines: 85 Here's a recent, true anecdote in the continuing "consumer electronics follies." (It's purely technical, and aimed mainly at readers who are heavily involved with electronics. Please observe the Followup: line and restrict posted responses, if any, to rec.audio.) First, background. An A/D or D/A converter does not become, say, "18-bit," simply by virtue of having 18 data wires. This undisputed and very fundamental point is sometimes missed, even by audio-engineering professionals, today (witness two or three major papers in JAES in the last couple of years). A true N-bit data converter meets or exceeds a whole set of implied analog specs that depend on N. For example, if an 18-bit A/D converter's low-order bits wiggle properly in response to small analog inputs, but wiggle much more from random noise, then no competent engineer would seriously consider it a true "18-bit" A/D, regardless of how eager the manufacturer's product-marketing group is to tenuously claim an 18-bit "resolution." A certain amount of cat-and- mouse of this kind lies in the background as high-resolution converters penetrate consumer markets through the vehicle of digital audio. Now, my anecdote. Not long ago at an international conference of the Audio Engineering Society (attended by some other usenet contributors too, as it happens), a competent and well-respected designer of analog/digital converter chips presented his design of a new "consumer" 18-bit A/D converter chip for the "high-end" market in DATs or similar recorders. The following impromptu and very frank exchange ensued when I got to the floor microphone (eagerly) at the end of his paper (this is taken from the taped record of the conference). (In case of the remote possibility that any usenet reader is unused to analog signal-fidelity measures in data converters, I have interpolated [in brackets] the implications, in "bits," of my remarks.) Q: "I'm a little confused by something, now. ... You quote a minimum 92 dB signal-to-noise ratio [corresponding to full performance in a FIFTEEN-bit converter]; ... 0.0025% THD-plus-noise, which is one part in 40,000 [or roughly the performance of a genuine SIXTEEN-bit device]. Can you explain to me in *precisely* what sense this is an `18-bit' A-to-D converter?" A: "I'd like to explain it from this standpoint. I think a lot of it comes from the input from the customers, who, quite frankly, use it for specsmanship reasons ... This product is geared strictly for a [consumer] application, and therefore you will find that the spec. numbers and `18 bits' don't really match up." Q: "Well, then, why not add a few more successive-approximation cycles, and call it `22' -- was it pin-limited? -- or `24' bits ..." [Acknowledgement of mutual understanding followed.] But the punchline is that this specsmanship doesn't matter, in the actual application! While the data converters do not really show 18-bit performance, they don't need to, since they are going into products with 16-bit data paths. So the consumers will get the "18-bit" converters they are screaming for, as well as the actual 16-bit performance (more or less) that their medium is limited to, without the cost of paying for (MUCH more expensive) true 18-bit data converters. A case study in efficient engineering, and everyone is happy. I wish the firm success. (Far more effective to the ear of the consumer -- if any consumers actually do use their ears -- than employing true 18-bit data converters would be other, cheaper, technical improvements, like decorrelating the quantization error using a sensible dithering scheme -- exit all the "granularity" and "audible distortion at low levels" effects -- or various improvements in AA and reconstruction filters. But these are not what consumers of "high-end" digital audio products are demanding.) There are certain other, less upbeat implications in this interchange, but I'll leave those to the astute reader. (I still wonder why not add more pins and call these converters 22 or 24 bits -- perhaps next year.) (P.S.: Please don't argue with me about the technical behavior of A/D's if you've been reading this newsgroup for less than a year.) Max Hauser / max@london.ee.cornell.edu (And various usenet paths with bangs in them, God only knows) -- "I'm oversampling by more than you are!" -- Stanley Lipshitz at the AES International Conference (taken well out of context). Copyright (c) 1989 by Max W. Hauser. All rights reserved.