Digital Processor Reviews

For logistical reasons, I measured different samples of the four Vivaldi components—transport, upsampler, master clock, D/A processor—from those auditioned by Michael Fremer. (The serial numbers can be found in the "Specifications" sidebar.) I examined the dCS Vivaldi system's electrical performance with Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see www.ap.com and the January 2008 "As We See It"). Both the Vivaldi DAC's and the Vivaldi Upsampler's performance via their USB ports was tested using my 2012-vintage Apple MacBook Pro.

Looking first at the Vivaldi SACD/CD transport, this offered almost the best error correction/concealment I have encountered. (The best is the Parasound Halo CD 1.) The Vivaldi played every track on the Pierre Verany Digital Test CD without glitches or muting until it reached track 38, which has 4mm gaps in the data spiral. The Transport recognized the pre-emphasis flag both on pressed CDs and on the CD-R with which the Parasound Halo CD 1 player had problems.

Turning to the Vivaldi DAC, Apple's USB Prober utility identified it as having the Product String "dCS Vivaldi DAC USB Audio 2" from "Data Conversion Systems Ltd"; the Vivaldi Upsampler was identified as "dCS Vivaldi UPS Audio Out." USB Prober confirmed that both products' USB inputs operated in the optimal isochronous asynchronous mode. The AES/EBU input locked successfully to data having sample rates up to 192kHz, with the dual-AES/EBU input handling DSD data and PCM data with sample rates up to 384kHz.

The maximum level at 1kHz from both balanced and single-ended outputs was 2.035V with the level set to "2V," 6.013V with it set to "6V." Both sets of outputs preserved absolute polarity (ie, were non-inverting), with the musical-note icon on the Settings screen right-side up. (The XLRs are wired with pin 2 hot.) The balanced output impedance was very low, at 2.3 ohms from 20Hz to 20kHz, including the series resistance of 6' of cable. The single-ended outputs had a higher source impedance, at a uniform 51 ohms, which is still low in absolute terms.

The Vivaldi DAC offers a choice of six reconstruction filters for data with sample rates of 44.1, 48, 176.4, and 192kHz, but only four for data sampled at 88.2 and 96kHz. Fig.1 shows the impulse response of Filter 1 with 44.1kHz data. The symmetrical ringing either side of the pulse maps the filter coefficients and reveals this filter to be a conventional linear-phase type. Filters 2–4 have increasingly shorter linear-phase impulse responses (fig.2 show Filter 4's impulse response), while Filter 5 (fig.3) is a minimum-phase type, with all the ringing following the pulse. Filter 6 (not shown) has similar time-domain behavior to Filter 1.

Fig.4 shows a wideband spectral analysis of the Vivaldi DAC's output with it set to Filter 1 and decoding 44.1kHz data representing a 19.1kHz tone at 0dBFS (cyan and blue traces) and white noise at –4dBFS (magenta and red, footnote1). The noise signal reveals that the filter rapidly rolls off the output above the audioband, reaching the noise floor at the Nyquist frequency, or half the sample rate. There is no trace of the image of the 19.1kHz tone at 25kHz (44.1–19.1kHz) above the 16-bit noise floor, and the second and third harmonics of the 19.1kHz tone lie at almost –110dBFS (0.0003%). Filters 2–4 offer increasingly slower rolloffs above the audioband, with Filter 4 (fig.5) allowing the 25kHz image to lie just 10dB below the level of the 19.1kHz tone and filling the audioband with spurious tones. It should be noted, however, that despite this slow rolloff, Filter 4 still doesn't offer the time-domain–optimized behavior of, for example, the famed Wadia Digimaster filter. Although it is a minimum-phase type, the Vivaldi DAC's Filter 5 (fig.6) offers identical image suppression to that of the linear-phase Filter 1, as does Filter 6.

Fig.7 shows the frequency response of Filter 1, with sample rates of 44.1, 96, and 192kHz. With each rate, the output is flat to just below the Nyquist frequency, with then a rapid rolloff. Filters 2–4 offer increasingly slower rolloffs with each sample rate (not shown), while Filter 5 offers a restricted bandwidth with 192kHz data, reaching –6dB at 40kHz (fig.8, blue and red traces). Filter 6 extends the bandwidth with 192kHz data, the –6dB point now lying at 55kHz (fig.9).

When the Vivaldi Upsampler is used to transcode 24-bit PCM data to DSD or the Vivaldi Transport feeds encrypted DSD data to it, the Vivaldi DAC offers a choice of four low-pass filters. Fig.10 reveals that the effect of these filters is to roll off the ultrasonic noise that results from the DSD format's noise shaping, with Filter 1 offering minimal rolloff (blue and red traces) and Filter 4, which dCS doesn't recommend for music playback, offering the most rolloff (green and gray). The traces in this graph were taken with original 44.1kHz data. Interestingly, when I used the Upsampler to transcode 24-bit/88.2kHz PCM data with the Vivaldi DAC set to DSD Filter 1, the traces were identical to the blue and red traces in fig.10. This suggests that while DSD encoding preserves the audioband resolution of 24-bit PCM, it actually has less resolution at ultrasonic frequencies.

The Vivaldi DAC's channel separation was superb, at >125dB in both directions below 1kHz (not shown), and still 118dB at 20kHz. The very low noise floor with 24-bit data can be seen in fig.11, with all power-supply–related spuriae at or below –135dB and the random noise components mainly stemming from the analyzer's A/D converter. Astonishing! With a dithered 1kHz tone at –90dBFS, increasing the bit depth from 16 (fig.12, cyan and magenta traces) to 24 (blue and red) dropped the noise floor by 24dB, indicating that the Vivaldi DAC has at least 20-bit resolution, which is the state of the art.

Linearity error with 24-bit data was nonexistent down to below –120dBFS, which, in conjunction with the low noise, allows it to perfectly reproduce an undithered 16-bit tone at exactly –90.31dBFS (fig.13). The waveform is perfectly symmetrical, the three DC voltage levels described by the data are perfectly resolved, and the time-symmetrical ringing from the linear-phase Filter 1 is clearly evident at the LSB transitions, even at this very low signal level. With Filter 5, the ringing clearly changes to the expected non–time-symmetrical type (fig.14), and with 24-bit undithered data, the result is a superbly well-defined sinewave (fig.15).

As good as the Vivaldi DAC's measured performance is in the digital domain, it is equally good in the analog domain. Fig.16 shows the output spectrum as the DAC drives a full-scale 50Hz tone at 6V into 600 ohms. The only distortion harmonics visible are the third, at –130dB (0.00003%) in both channels, and the second, at –126dB (0.00005%) in the left channel (blue trace). Intermodulation distortion was also vanishingly low, although, as expected from the white noise/19.1kHz spectra discussed earlier, the rejection of the ultrasonic images resulting from an equal mix of 19 and 20kHz tones sampled at 44.1kHz depended on the reconstruction filter used. Filter 1 (fig.17) gave almost total suppression, Filter 4 the least suppression (fig.18), with again a lot of spuriae dumped back into the audioband.

It came as no surprise that the Vivaldi DAC, fed directly or via the Upsampler, offered superb suppression of jitter. With a 16-bit version of the Miller-Dunn J-Test data fed to the DAC's AES/EBU input, all that could be seen were the odd-order harmonics of the LSB-level tone, all at the correct level (not shown). With 24-bit J-Test data, a single pair of signal-related sidebands at ±229Hz is visible at –142dB (fig.19), though low-level sideband pairs of unknown origin can also be seen at ±71 and ±142Hz. With 24-bit data fed to the Upsampler and transcoded to DSD before being sent to the Vivaldi DAC, the central spectral peak that represents the 11.025kHz tone in fig.19 widened very slightly at its base and reduced the level of the sidebands at ±229Hz (fig.20).

With 24-bit data fed to the DAC via USB from my MacBook Pro (fig.21), there was a little less spectral spreading of the 11.025kHz tone than with AES/EBU, and the signal-related sidebands have dropped even further. The sidebands at ±142Hz are still evident, however. Again, connecting the MacBook's USB output to the Upsampler and transcoding to DSD before feeding the data to the DAC completely eliminated the sidebands at ±229Hz and broadened the central spike a little, but reintroduced the low-level sideband pair at ±71Hz (not shown).

I repeated the USB-input jitter tests, clocking both the Upsampler and the DAC from the Vivaldi Clock. Whether it was the DAC fed 44.1kHz J-Test data via USB or fed 44.1kHz data upsampled to DSD by the Upsampler, I didn't find any significant measured differences from the DAC and Upsampler's internal clocks. But from my auditioning, I did find the Vivaldi Clock to add a small but noticeable improvement in sound quality, as I had done in my March 2005 review of the dCS Verona Clock.

Overall, the dCS Vivaldi measured superbly well. When I measured the company's earlier four-box system, the Scarlatti, in 2009, I concluded that it offered "state-of-the-art measured performance." The Vivaldi improves on the Scarlatti's performance in almost every way. Wow!

Then I took a listen to the complete Vivaldi system. (See my Pass Labs review elsewhere in this issue for a list of the ancillary components used.) Again wow! Without any doubt, this was the best digital playback I have experienced. While there are digital processors that offer measured performance almost as good as the Vivaldi—the Electrocompaniet ECD-2, for example, and MSB's Diamond DAC IV or even NAD's M51—there was an increased sense of ease to the Vivaldi's sound quality, particularly when the Upsampler transcoded the data to DSD, that proved addictive. In this respect, it was even better than the MSB it replaced in my system. For a third time: Wow!—John Atkinson