dCS Verona Master Clock Measurements
Full sets of measurements of the dCS Elgar Plus playing back CDs and of the upsampling effect of the dCS Purcell were published in their original reviews in July 1997 and January 2001, respectively. Measurements of the Verdi SACD transport driving the Elgar Plus via dCS's proprietary IEEE1394 FireWire link accompanied Michael Fremer's review in April 2003.
To examine the effect of the Verona on the dCS system, I first looked at the Elgar's jitter performance with the Miller Audio Research Jitter Analyzer while the DAC decoded CD data from the Verdi, upsampled to a DSD datastream with the Purcell. The transport was word-clock–"slaved" to the DAC, dCS's recommended mode. The resulting narrowband spectrum of the Elgar's analog output, plotted to a different, "zoomed-in" scale compared with my usual graphs, is shown in fig.1. The absolute jitter level was a moderately low 395 picoseconds peak–peak (footnote 1).
Fig.1 dCS Elgar-Purcell-Verdi, high-resolution jitter spectrum of analog output signal with Elgar as master word-clock source (CD data, 11.025kHz at –6dBFS sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±1.5kHz.
Although the data-related sidebands (red numeric markers) are all almost at the residual level of the test signal—it is an unfortunate coincidence that when Paul Miller designed this test system, he chose a low-frequency squarewave whose harmonics are spaced symmetrically around the 11.025kHz diagnostic tone—four pairs of sidebands contribute most of the measured jitter. These lie at ±15.1Hz (purple 1), ±31.7Hz (purple 2), ±913Hz (purple 6), and ±1246Hz (purple 9). I don't have a clue what causes these sidebands to exist, but the two low-frequency pairs seem ubiquitous in CD players.
For comparison, fig.2 shows the Elgar's output spectrum when it, the Verdi, and the Purcell are all word-clock–slaved to the Verona with the latter's dither switched off. The overall jitter level has dropped to 312ps p–p. The data-related sidebands remain just above the residual level, and though the sidebands at ±15.1Hz and ±31.7Hz might still be present, these are obscured by the broader "skirts" around the central peak that represents the 11.025kHz tone. This behavior is probably due to the presence of some low-frequency random jitter, but note that a low-level sideband pair at ±45.4Hz makes an appearance.
Fig.2 dCS Elgar-Purcell-Verdi, high-resolution jitter spectrum of analog output signal with undithered Verona word-clock master (CD data, 11.025kHz at –6dBFS sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±1.5kHz.
There is also a pair of low-level sidebands now visible at ±120Hz (blue "2" markers), which is, not coincidentally, the full-wave–rectified power-supply frequency. The Verona adds another ground-return path, which apparently allows a minuscule amount of 120Hz hum to affect the conversion. While the other jitter-related sidebands are still present (purple numeric markers), these have had their energy shifted a little higher in frequency.
Fig.3 shows what happens when the Verona's word-clock dither is switched on. The jitter level has risen slightly, to 342ps. The skirts around the central peak have broadened a little further, obscuring the ±45.4Hz sidebands. Though the sidebands at ±120Hz are still evident, those at ±230Hz, which are data-related (red "2" markers), have been reduced to the residual level. Conversely, the highest-frequency sidebands have increased in level.
Fig.3 dCS Elgar-Purcell-Verdi, high-resolution jitter spectrum of analog output signal with dithered Verona word-clock master (CD data, 11.025kHz at –6dBFS sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±1.5kHz.
I offer these measured results not so much to explain what I heard when I used the Verona—with the exception of the spreading of the central peak, from which I have heard a similar improvement in sound quality in the past—but to show that something is going on that is reflected in the Elgar's analog output both when the Verona is used, and whether it is used with or without word-clock dither.
The final measured result involves the Metric Halo ULN-2 microphone preamplifier and A/D converter that I used for the main pickup on my 2004 recording of Cantus performing Christmas music, slaved to an 88.2kHz AES/EBU link from a dCS 904 A/D converter. While the ULN-2 would lock on to the Verona's word-clock output, this was only when the dither was switched off. The ULN-2 would lock to the Verona's S/PDIF output whether it was dithered or not—the S/PDIF output consists of data representing "digital black"—so I examined its performance under those conditions.
The test setup was to drive the Metric Halo's microphone input with a balanced high-level 20kHz tone sourced from my Audio Precision System One. The sample rate was set at 44.1kHz, and the input level controls were set so that the converter's AES/EBU digital output gave 24-bit data representing the 20kHz tone at –3dBFS. The combination of a high-level tone with a high frequency close to Nyquist (footnote 2) is an unrealistically horrible test for an A/D converter—the resultant digital-domain spectrum produced by the ULN-2 when set to its master clock mode (fig.4, red trace) suffers from modulation of the noise floor around the 20kHz tone, some sidebands at ±1kHz and ±1.9kHz, and the introduction of some discrete tones. (The FFT analyzer was that included with Adobe Audition, set to 65,356 points with a Blackmann-Harris window; 30s of data were analyzed.) Fortunately, almost all the enharmonic garbage lies at or below –110dB. However, the situation didn't improve when the ULN-2 was clocked from the Verona, using the latter's undithered word clock.
To my surprise, the ULN-2 behaved a lot better when clocked from the Verona using the latter's dithered S/PDIF word signal (fig.4, blue trace, obtained under the same conditions). Both noise modulation and discrete tones are effectively banished, though the noise floor below 10kHz is up to 6dB higher in level. I had assumed that this change in the noise floor was due to the dither the Verona was applying to the word-clock signal. However, repeating the spectral analysis without dither gave an identical result. This is plotted in green in fig.4, but you can't see it because it is exactly overlaid by the dithered spectrum.
Fig.4 Metric ULN-2, high-resolution spectrum of digital output signal when fed 20kHz at –3dB in master clock mode (red) and when clocked by dCS Verona's S/PDIF word-clock signal when dithered (blue) and undithered (green).
General conclusions shouldn't be drawn from this test, as A/D converters differ considerably in how they handle reference word-clock signals, and an almost full-scale 20kHz signal will never be met in real life. But it does make me glad I ended up clocking the ULN-2 with a reference AES/EBU clock signal derived from that of the Verona for my 2004 Cantus sessions. Again, it shows that changes in clocks produce changes in digital audio data that are measurable and hence, possibly, audible.—John Atkinson
Footnote 1: For reasons unknown, all these measured jitter levels are about twice what I measured for the Elgar and Verdi two years ago. They should therefore be considered as relative rather than absolute values.
Footnote 2: The Nyquist frequency is exactly half the sample rate.