The Law of Averages

The Compact Disc clearly hasn't read the script. At a time when, in the autumn of its commercial life, the format is supposed to be stepping aside to allow younger blood to succeed it, CD has instead in recent years enjoyed something of a revival in audiophile opinion. While SACD and DVD-Audio, rather strutting and fretting their hour upon the stage, are doing more plain fretting than anything, the best in CD sound quality has improved sufficiently for some to question whether we need the new media at all.

I think all the hi-rez and multichannel naysayers are misguided—and those who believe two-channel stereo to be some kind of theoretical exemplar plain barmy—but there's no denying that developments such as upsampling have, for reasons I'm not certain anyone really understands, upped what 16-bit/44.1kHz digital audio can achieve.

What I'm going to tell you about here is along broadly these lines, although in this case the change is to the data recorded on the disc rather than to the hardware that plays it. In one sense, that makes this story significantly less universally interesting, because the technique has been used on only a small number of discs, all of them of classical music. Moreover, the idea I am introducing is probably applicable only to this type of program, and not even all of that. But because it calls into question some of the core assumptions inherent in digital audio, it deserves to be aired even if its impact on your particular listening pleasure may be small to nonexistent.

Tony Faulkner: name mean anything? If you are a devotee of indie classical label Hyperion, for instance, and an absorber of disc small print, probably it will. Tony is, and has been for the past 20 years or so, one of the UK's top freelance recording engineers. (There was a period a few years back when he briefly became a corporate man, but we won't go into that. Suffice it to say it didn't last.) Moreover, he was one of the first—the first—to embrace high-rate, high-resolution digital recording when the technology first became available. I recall lending him one of the early Pioneer double-speed (16/96) DAT recorders in 1993 while I was reviewing it, to make some comparison recordings between 48kHz and 96kHz sampling at his recording sessions, an experience that I think acted as a catalyst. Tony was no more enthusiastic about DAT as a mastering medium than I was, but soon thereafter he equipped himself with open-reel digital recorders that would allow him to capture those improvements he'd heard via the Pioneer.

Even when there is no prospect of a hi-rez release—which is normally the case—Tony habitually records at 176.4kHz so that future options are kept open, something his clients may have cause to thank him for in years to come. Multiples of 44.1kHz are preferred to 96kHz or 192kHz sampling rates because the downsampling (we should properly call it decimation) for CD mastering is simpler and less prone to the sonic side-effects that typically afflict asynchronous downsampling (ie, where the higher sampling rate is not an integer multiple of the lower).

For many years, Tony generated 44.1kHz CD masters from his high-rate originals using conventional downsampling techniques. But three years ago he began, without any fuss or fanfare, to use a different method—one that breaks the "rules." Here, in his own words, is how, more by happenstance than by contrivance, it came about:

"We've recorded everything 176.4kHz for about five years, so that if anybody really gets their act together with a high-density format, we're ready for it. But editing was an issue for people who didn't have sample-rate converters. If you want to create an archive of 176.4kHz masters, then you record and edit them at 176.4kHz, but while you edit you've got to be able to hear them. Here I have the luxury of a dCS sample-rate converter, which I can put in the output of the editor and hook up to my normal D-to-A. But when I subcontract editing out or it goes to a producer who does their own editing at home, which happens a lot, I've got to give them something they can work with. So I started delivering what looks like eight-track 24-bit/44.1kHz, with the first four tracks being adjacent samples of the 176.4kHz original's left channel and the other four tracks the same for the right channel.

"For editing purposes you can listen to the first track as the left channel and the fifth track as the right channel, so you're doing a crude decimation with no [low-pass] filter. The sound's a bit bright—particularly on string quartets and piano, where you've brought the average level up, you can hear that telltale zizz of aliasing—but it's okay for editing purposes.

"I got a bit bored of this and thought I'd average the samples from each set of four tracks, dither it, and call that 24-bit/44.1kHz instead. Initially, I just did it for convenience, but when I listened to it I thought, 'That sounds nice.' I'm so used to hearing what 44.1kHz sampling sounds like—not just dull, there's a lack of impact and a dead quality that audiophiles have always hated about CD. Using the averaging, there was less of this.

"I had two titles on the computer waiting to be edited, the first of which was [Berlioz's] Symphonie Fantastique with Sir Colin Davis (LSO Live LSO0007). There had been a fair amount of whinging from critics, players, and others about the LSO Live recordings because the Barbican Hall is a bit on the dead side, and this was compounded by the deadness of CD. So I thought, 'I'll put this one out with no filters, using my averaging technique, and see what everybody thinks.' I listened very critically to it first, burnt a copy for the office, burnt a copy for the producer, and everybody came back and said, 'Wow, this sounds nice.'

"The reviews were very favorable too, despite this being yet another Symphonie Fantastique and, not only that, yet another from this conductor. They didn't all say, 'Gosh, I can't hear 100dB per octave anti-alias filtering.' What they did say was that it sounded exciting, and that this is refreshing after some of the previous ones, which we thought a little lackluster sonically.

"So I then tried the same on a solo piano recording of Marc-André Hamelin. I could hear more aliasing with that, but still there was more excitement to the performance using the averaging technique. I didn't call it 4D or anything to make out that it was different, I just hoped people would like it. If they didn't, I could always stick my hands up, apologize, say we'll have to remaster it and I'll pay. But that hasn't happened yet."

The first wind I got that Tony was up to something unusual came at the Audio Engineering Society's 17th UK Conference, in April 2002, where he delivered a paper entitled "Discs: New Formats Bring Optical Media to Life." Actually, he didn't deliver his written paper at all: he busked it by offering some thoughts from the podium and then inviting comments from the audience, which I suspect caused many to miss a section of the text that made my eyes light up:

"With regular PCM and DSD we have done listening tests with musicians on a regular basis, and consistently they object to the sound quality of some, or in some cases all steep roll-off low-pass filters. Some musicians even prefer the sound of fairly serious aliasing to that of over-steep filters. One pianist said to me: 'This is very interesting—you are asking me to choose between my instrument sounding as though it is covered with blankets and in a carpeted room, or else in a live space only with a more metallic hardness. There is no choice—I prefer the nails to the blankets.' "

Footnote: Keith Howard, for many years editor of the UK magazine Hi-Fi Answers, is currently Consultant technical editor, Hi-Fi News, Technical consultant, Autocar, Special contributor, Motor Sport, and a contributor to Racecar Engineering.