Sonic Frontiers SFD-2 D/A processor JA on HDCD

Sidebar 7: JA on HDCD

The obvious comparison for a new technology such as HDCD is to audition the same recording in HDCD and non-HDCD formats through the same D/A processor, so that you can isolate the effect on sound quality due to the HDCD process alone. But this is hard to do. You can audition an HDCD disc through the SFD-2 Mk.II processor, which will give you the full HDCD sound quality, but you can't play the HDCD disc through the SFD-2 Mk.II with the HDCD decoding disabled. You can play the HDCD disc through a non-HDCD D/A, but then the inherent sonic differences between that processor and the SFD-2 will confuse things.

"Aha!" I thought. Why not hook up two datastreams to the SFD-2 Mk.II: one direct from the Mark Levinson No.31, to give an HDCD source; the other via the Audio Alchemy DTI•Pro, set to redither at the 16-bit level? As the DTI•Pro will change each data word's LSB, which is where the HDCD code resides, I would then have a non-HDCD datastream describing exactly the same music.

So that's what I did. The auditioning was confusing, however, as sometimes the improvement from HDCD decoding didn't seem nearly as great as I had expected. The HDCD sound, particularly on Reference Recordings' excellent Testament and Rutter Requiem discs (RR-49CD and RR-57CD, respectively), was always superbly natural. The non-HDCD playback, however, didn't always sound much worse, and sometimes sounded quite different.

One interfering factor is that non-HDCD playback changes the sound of the recording. The sound of HDCD discs when played back non-decoded is very gently compressed. Other than on Dick Hyman Plays Duke Ellington (Reference RR-50DCD), this is not sufficient to give rise to noise "pumping" artifacts—and then only between cuts—but it does make room reverberation more apparent, and slightly brings up the music's average level. The latter makes the music sound "louder," even if the peak levels are matched—meaning that there's sufficient difference between HDCD and non-HDCD playback to confuse the auditioning.

When I discussed this test with UltraAnalog's Richard Powers and Pacific Microsonics' Michael Ritter at the 1995 WCES, both pointed out that my intended comparison would be invalid. First, the changes wrought by the Audio Alchemy device—DSP and jitter reduction—would themselves change the sound, introducing a significant additional variable. Second, Mike Ritter explained to me that the altered datastream from the DTI•Pro, as it originally was HDCD-encoded, would not be treated by the PMD100 digital filter in the same manner as a real non-HDCD datastream. "The chip also does other things," was his enigmatic statement; but I didn't understand exactly what those other things were. Certainly, if true, this would be yet another variable to interfere with the comparison.

It turns out that the only real comparison I could carry out was to use the one commercial recording that has been issued in both HDCD and non-HDCD formats: Dick Hyman's Duke Ellington homage (Reference RR-50CD and RR-50DCD, respectively). There are other differences, in that the HDCD disc was recorded directly to the glass CD master and pressed by Superior Disc in Tokyo using a gold reflective layer, while the non-HDCD disc was mastered conventionally and pressed by Disc Mfg., Inc. using a normal aluminum reflective layer. However, the primary difference between the two CDs is that one encoded the sound of the piano with the prototype HDCD encoder, the other used Keith Johnson's highly modified Sony PCM-701ES. The comparison is therefore between the entire HDCD chain and a conventional, if high-quality, 16-bit chain.

There's still one variable which will affect the comparison: As totally separate ADCs were used, and a reference tone was apparently not recorded on both master tapes, the possibility exists of there being a level difference between the two recordings. I therefore checked this both in the digital domain using an AES/EBU-input Dorrough peak/average-reading meter, and in the analog domain using the peak-hold function of the AudioControl Industrial SA-3050A spectrum analyzer. In the analog domain, the peak 1/3-octave band levels achieved during the first 1:02 of track 3 of the Ellington disc were within 1dB from aluminum to gold discs, with perhaps the non-HDCD disc being a fraction of a dB louder across the band. In the digital domain, track 3's opening chords reached peak levels of -10dBFS (L) and -14dBFS (R), with the average level during the song's chorus cruising around the -20dBFS mark.

In fact, the only appreciable measurable difference between the two discs involved the noise floor at the start of the disc: Whereas the noise on the regular disc reached peak levels of -54dBFS (L) and -57dBFS (R), both channels of the HDCD disc featured a background level of -42dBFS. What was fascinating was that I couldn't hear any noise. Checking with the spectrum analyzer revealed why: the noise was above my range of hearing. Presumably, the HDCD's encryption of the extra information is in the form of apparently random noise near the Nyquist frequency.

Using the Ellington disc, it was possible to hear that, as good as the non-HDCD version sounded, it was a shadow of the HDCD version. Tonally, I was hard put to hear any difference—it was obviously the same piano. But in its ability to reproduce the sound of the space surrounding the piano, the gold HDCD disc went far beyond the aluminum disc. Instead of rendering a rather anonymous-sounding acoustic, the HDCD process seemed to allow me to "hear the walls" of the recording hall, yet without the reverberation swallowing up the direct sound and making the recording sound too "bathroomy." Instead, the more apparent coherence of the HDCD disc's reverberant sound made the piano sound more "modeled," in the sense of revealing the solid nature of the instrument. By comparison, the non-HDCD piano sounded more flat, more one-dimensional. Once I got used to the sound of the HDCD piano, it was hard to go back to the ordinary one. (And to those critics who feel all such differences are due to frequency response and amplitude differences, note that my measurements indicate the spectra of the two recordings to be identical, and that the ordinary piano was slightly louder—which, the conventional wisdom would have it, would make me prefer it in such comparisons.)

In my experience, this is exactly the kind of difference I expect between good analog and good digital recording. The former has an image palpability which the latter, until recently, has not been able to achieve. And then, only completely in the context of 20-bit storage media. (Noise-shaping redithering processors—such as Sony's Super Bit Mapping processor, the Meridian 618/518, and the Apogee UV-22—allow a good deal, but not all, of that quality to make its way onto CD.)

In the context of a good high-end system, the HDCD process offers a genuine improvement in sound quality. It enables the CD to burst out from the straitjacket imposed by the CD standard's limited data window while ensuring compatibility with non-HDCD playback systems. But making the source better will do nothing about problems elsewhere in the system or listening room. At the recent WCES, some of the best sound I heard at the Show was in rooms that used the Sonic Frontiers SFD-2 Mk.II to play HDCD-encoded discs. But so was some of the worst sound. And there remains the fact that, once you've played all your Reference Recordings CDs twice over, you're left thirsting to play some of the music you bought because of the music.

I have this fear that HDCD is two years too late to make the impact on the market it deserves. (The recent announcement by Sony and others of CD-like video media with much larger data capacities—3.7 gigabytes—confuses the issue, though their commercial introduction is probably two years away.) True, 16 hardware manufacturers had signed licenses to manufacturer HDCD decoders as of December 1994; but at the time of writing, HDCD encoders are still weeks away from getting into the hands of record companies other than Reference Recordings. A ratio of 16 hardware licensees to only one software licensee is the wrong way around for instant commercial success, I believe (though Warner Bros. has just announced that their reissues of the Neil Young catalog will be HDCD-encoded). As both RH and I have found that the PMD100 chip is one heck of a fine digital filter, it would be ironic if that, rather than the fact that it offered HDCD decoding, became the primary reason for its sales success. The next year is going to be very interesting.—John Atkinson