dCS Verdi SACD transport, Purcell D/D converter, Elgar Plus D/A converter Measurements

Sidebar 3: Measurements

As full sets of measurements of the dCS Elgar Plus playing back CD data and of the upsampling effect of the dCS Purcell were published in their original reviews (July 1997 and February 2001, respectively), I refer readers to the Web reprints of those reviews, available in our online archives. For this review, I looked only at the Elgar's measured behavior fed SACD data from the Verdi transport via dCS's proprietary IEEE1394 link. As recommended in the comprehensive manual, the Verdi was slaved to the Elgar. The test signals used were those on Sony's "provisional" Test SACD.

With its analog output switch set to High, the Elgar's maximum output at 1kHz was 6.048V balanced and 5.923V unbalanced. The Low setting gave 2.019V and 1.977V, respectively. This was with SACD playback. CD playback gave figures 1.3dB lower, which will be audible in direct comparisons not so much as a level difference but as less vitality and drive to the sound of CDs. The Verdi-Elgar's frequency response for SACD playback is shown in fig.1. Perfectly flat in the audioband, it starts to roll off above 30kHz, reaching -1dB just above 60kHz. The output is down 11.5dB at 100kHz, and continues rolling off above that frequency.

Fig.1 dCS Verdi & Elgar Plus, SACD frequency response at -3dBFS into 100k ohms (right channel dashed, 1dB/vertical div.).

This rolloff is to prevent potentially harmful levels of ultrasonic energy being fed to the playback system. As explained in the DSD sidebar in our review of the pioneering Sony SCD-1 SACD player, the DSD encoding used with the SACD medium results in very high levels of random noise extending from just above the audioband up to the region of the 2.82MHz sampling frequency. The residue of this noise, following the Elgar's low-pass filtering, can be seen in fig.2, well above the noise level of the 24-bit LPCM signal, though this does show a much smaller rise above 100kHz. The DSD peaks at around -42dBFS between 80kHz and 100kHz, meaning that the SACD medium has only about 7 bits of resolution in this region. In the audioband below 10kHz, however, the resolution is much better than the CD's 16 bits, evidenced by the Elgar's noise floor reaching almost -140dBFS at low frequencies. However, as the Elgar's 24-bit LPCM noisefloor in this region is even lower in level, this appears to demonstrate that DSD is not quite up to 24-bit PCM encoding in terms of ultimate resolution.

Fig.2 dCS Verdi & Elgar Plus, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS (top) and -120dBFS (middle), DSD data; and of 1kHz tone at -120dBFS, 24-bit LPCM data (bottom). (Right channel dashed.)

Fig.2 graph shows 1/3-octave spectral analyses of the Elgar's analog noise floor while it decoded DSD data representing dithered tones at -90dBFS and -120BFS, as well as the 24-bit LPCM equivalent of the latter. The reproduced levels of all three tones are correct, implying excellent linearity, and no harmonic spuriae can be seen. However, with the -120dBFS DSD tone, a second peak can be seen between 600Hz and 700Hz. This was repeatable, and was also absent from the LPCM spectrum. It might well, therefore, be due to an idle tone of some kind associated with the DSD encoding. It also looks from this graph that some power-supply spuriae are present in the right-channel output. However, these are still 130dB down from peak level—a good working definition of "negligible"!

Fig.3 plots the Elgar's linearity error for SACD playback, using the 1kHz spot tones on the Sony Test SACD. The measured level error is effectively zero down almost to -120dBFS, which is superb. It rises below that level due to the increasing contribution of noise. Even so, it is still just 2.5dB at -130dBFS!

Fig.3 dCS Verdi & Elgar Plus, left-channel departure from linearity, DSD data (2dB/vertical div.).

The Elgar Plus also offered superbly low levels of harmonic distortion. Fig.4 shows the spectrum of a full-scale 1kHz tone played back off the Sony SACD. The THD (the actual sum of the harmonics, not including noise) was 0.0006% in the left channel and 0.0009% in the right. The latter was affected by a higher level of third harmonic, which was -101.7dB in the right channel compared with -112.4dB in the left. The difference is inconsequential. There is a 60Hz component present at -90dB, but this was due to a ground loop between the dCS components and the computer that acts as the platform for some of my measuring gear that I couldn't eliminate.

Fig.4 dCS Verdi & Elgar Plus, spectrum of 1kHz sinewave, DC-10kHz, at 0dBFS into 8k ohms (DSD data, linear frequency scale).

I measure the effect of word-clock jitter by using the Miller Audio Research Analyzer to perform narrowband spectral analysis on the analog output of the device under test while it decodes an analytical signal developed by the late Julian Dunn (footnote 1), founder of PrismSound. This consists of a high-level sinewave at one quarter the sampling frequency (see March 2003, p.97), over which has been laid a low-frequency squarewave at the LSB level.

I don't have this signal available on SACD, but Sony's Test SACD does have a high-level tone at 11.025kHz. It lacks the low-frequency squarewave, however, and so will not excite data-related word-clock jitter. The black trace in fig.5 shows a high-resolution spectrum of this tone, played back on the Verdi-Elgar combo. The actual noise floor is at or below -130dBFS, which is what I was expecting. What I was not expecting were some sideband pairs, at frequencies of ۰Hz (purple "1" numeric markers), 䕙Hz (purple "5"), 䖄Hz (purple "6"), and 2.45kHz (purple "17"). However, even with these sidebands, the calculated level of jitter was just 147.8 picoseconds, which is exceedingly low.

Fig.5 dCS Verdi & Elgar Plus, high-resolution jitter spectrum of analog output signal, SACD playback (11.025kHz at -6dBFS). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. (Grayed-out trace, 11.025kHz at -6dBFS sampled at 44.1kHz with LSB toggled at 229Hz, CD playback upsampled to DSD by dCS Purcell.)

The grayed-out trace in fig.5 is a similar analysis of the Elgar's output while it decoded the Dunn signal, which was stored on CD, played back on the Verdi, and upsampled to DSD by the dCS Purcell. The noise floor is a little higher in level. While the jitter level has risen to 185ps, this is still excellent. The sidebands at 䖄Hz are still visible, if a little higher in level, but those at 2.54kHz have disappeared. What is especially commendable, however, is the very low level of data-related jitter. Repeating this measurement, but with the Purcell used to upsample the CD data to 192kHz and driving the Elgar with a dual AES/EBU link, gave a higher jitter level, 580ps (not shown). Peculiarly, while there were more data-related sidebands present, most of the extra jitter was in the form of strong sidebands in the 200Hz region.

Driven by the Verdi transport, dCS's Elgar Plus offers state-of-the-art playback, not only of SACDs but also of CDs. In addition, all three dCS products are effectively future-proof. It is hard to see how digital technology can get any better than this.—John Atkinson

Footnote 1: I was saddened to learn of Dunn's untimely passing, on Thursday, January 23, 2003, at Addenbrooke's Hospital in Cambridge, England. One of digital audio's creative thinkers, Julian had been ill with leukemia for some time.—John Atkinson
US distributor: Audiophile Systems
8709 Castle Park Drive
Indianapolis, IN 46256
(888) 272-2658