Accuphase DP-100 Super Audio CD transport & DC-101 Digital Processor Measurements part 3

Of much more significance in this graph is the very high level of ultrasonic noise resulting from the aggressive noiseshaping used by the DSD process to achieve such a wide audioband dynamic range from a 1-bit encoding system. The noise appears to peak at 75kHz in fig.11, but this is the result of the player's ultrasonic rolloff (seen in fig.10).

The method of analog analysis used to capture this spectrum possibly exaggerates the impact of DSD's HF noise within the audioband. Fig.12 therefore shows an FFT-derived spectrum taken while the Accuphase decoded DSD data representing a 1kHz tone at -60dBFS. A smidgen of second harmonic can be seen at -120dB, while the noise floor remains "white" (ie, flat with this linear frequency-scaled graph) up to 20kHz. Fig.13, taken by digitizing the Accuphase's analog outputs at 96kHz while the DP-100 played back a full-scale DSD-encoded 1kHz tone, confirms this behavior. In this graph, the noise floor continues to rise to 45kHz, above which it is rolled off by the Digital Audio Labs CardDeluxe A/D converter's antialiasing filter. Again, some very-low-level, low-order harmonics can be seen, but these will be at least partly due to the CardDeluxe's ADC.

Fig.12 Accuphase DC-101, SACD data, FFT-derived spectrum of dithered 1kHz tone at -60dBFS (linear frequency scale, 0Hz-24kHz, 30dB/vertical div.).

Fig.13 Accuphase DC-101, SACD data, FFT-derived spectrum of dithered 1kHz tone at -0dBFS, digitized at 96kHz with a CardDeluxe 24-bit ADC (linear frequency scale, 0Hz-48kHz, 6dB/vertical div., left channel cyan, right channel magenta).

As I had the Accuphase hooked up to the CardDeluxe's analog inputs, I took the opportunity to digitize some SACD discs, in order to follow up on my spectral-analysis work on wide-bandwidth PCM recordings (see Stereophile, October 2000, pp.63-71, and January 2001, pp.175-179). Fig.14 shows a 3D spectral plot of the CD-layer track of Mighty Sam McClain's "Too Proud" (from the SACD Soul Survivor, AudioQuest Music AQ-SACD053), again digitized at a 96kHz sample rate with the CardDeluxe. The vertical axis in this graph is frequency, extending up to 48kHz; the horizontal axis is time; and the color of the spectral components corresponds to amplitude, with white, then yellow, being the highest in level, and dark blue, then black, the lowest.

Fig.14 Spectrogram of Mighty Sam McClain's "Too Proud," from Soul Survivor (AudioQuest Music AQ-SACD053). Original SACD CD-layer data digitized at 24 bits/96kHz.

As expected, there is no spectral content above the CD's Nyquist Frequency of 22.05kHz, but there are plenty of spectral lines evident that have HF content up to that limit, these mainly due to cymbals and electric guitar. Performing an identical analysis on the SACD-layer track gave the spectral plot shown in fig.15. The musical content can be seen to extend well above 22kHz, showing that the DSD encoding does capture ultrasonic information, as claimed. But look at what happens above 40kHz in this graph: DSD's noiseshaping energy makes an appearance.

Fig.15 Spectrogram of Mighty Sam McClain's "Too Proud," from Soul Survivor (AudioQuest Music AQ-SACD053). Original SACD DSD-layer data digitized at 24 bits/96kHz.

When I used the Accuphase to audition the SACD- and CD-layer versions of the Mighty Sam McClain track, even though the mastering from the analog original seems to have been identical, the SACD layer sounded very much better. It was more like "analog" in that images within the soundstage were better fleshed out, there was a total absence of treble glare and grit, and the bass sounded both better defined and with greater LF extension.

What I found very interesting is that when I compared the PCM WAV files used to generate figs.14 and 15, using Sennheiser HD580 headphones driven from the monitor output of the RME Digi96/8 Pro soundcard I reviewed in January, the 24/96 WAV version made of the decoded DSD-layer data sounded much better than the 24/96 WAV made from the decoded CD-layer data. As best as I could tell—peak level, audioband spectral content—the recordings are identical other than the presence of information above 22.05kHz in the DSD-sourced version. Given that the dynamic headphones I used roll off above the audioband, and that I can't hear above 16.5kHz anyway, I suspect that a hypothesis advanced by Doug Rife in the December 2000 issue's "Letters" (p.12; see also Doug's paper on the subject) has merit: that the presence of random ultrasonic energy—as is present in the DSD-sourced PCM WAV data—will linearize a PCM D/A converter.

Back to the Accuphase DC-101/DP-100: They performed superbly well on the test bench with respect to both CD and SACD replay.—John Atkinson

US distributor: Axiss Distribution
17800 South Main Street, Suite 109
Gardena, CA 90248
(310) 329-0187