AURALiC VEGA D/A processor Measurements
I used 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") to examine the Auralic Vega's measured behavior. I used my 2012-vintage Apple MacBook Pro to examine the processor's performance via its USB port. All measurements were taken with the Vega's volume control, which operates in accurate 0.5dB steps, set to its maximum of "100."
Apple's USB Prober utility identified the Vega as having the Product String "AURALiC USB Audio 2.0" and the Manufacturer String "AURALiC," and confirmed that the Vega operates in the optimal isochronous asynchronous mode. The Vega's AES/EBU and S/PDIF inputs successfully locked to datastreams with sample rates up to 192kHz, including TosLink, which is usually restricted to 96kHz. The maximum output voltage was the same from both balanced and unbalanced outputs, at 4.37V, this 6.8dB higher than the CD standard's 2.0V. With the Phase control set to Normal, both sets of outputs preserved absolute polarity (ie, were non-inverting). The output impedance from the balanced jacks was an extraordinarily low 0.6 ohm at all audio frequencies. It was the specified 50 ohms from the unbalanced jacks, which is still usefully low.
The four filters offered by the Vega for playback of PCM recordings differed in behavior in both the time and frequency domains. Fig.1 shows the impulse response with 44.1kHz data of Filter Mode 1. It is a conventional time-symmetrical, finite-impulse-response type, with the "ringing" mapping the filter's coefficients. Modes 2 and 3 feature successively shorter amounts of pre- and post-ringing (fig.2), while Mode 4, which I preferred overall, is a minimum-phase type, with all the ringing following the main pulse (fig.3). This ringing, however, neither lasts as long nor is as well developed as with other minimum-phase reconstruction filters (see, for example, fig.2 here).
To highlight the differences between these filters in the frequency domain, I use a representation suggested to me by Jürgen Reis of MBL, in which the device under test decodes first 44.1kHz data representing white noise, then 44.1kHz data representing a full-scale tone at 19.1kHz. Fig.4, for example, shows the resulting spectrum with these signals with the Vega's Mode 1 filter. The white-noise spectrum reveals that the Vega's output rolls off rapidly above 21kHz; as a result, the sampling image of the 19.1kHz tone at 25kHz (blue and magenta traces) is completely suppressed. Note also the very low levels of the distortion components of the 19.1kHz tone in this graph. Filter Mode 2 is almost identical (not shown), but Mode 3 (fig.5) has a much slower rate of ultrasonic rolloff. Consequently, the 25kHz image of the high-level tone is reduced by just 24dB. Mode 4 offered another 6dB of image rejection (fig.6), but the still-slow rolloff correlated with the smaller amount of ringing in seen in fig.3.
Fig.7 is a conventional frequency-response plot, taken with Filter Mode 1 and sample rates of 44.1, 96, and 384kHz. The channel levels match to within 0.05dB, and the shape of the response is the same with all three sample rates: flat almost to 20kHz, with then a gentle rolloff above the audioband broken by a sharp cutoff just below each of the three lower Nyquist frequencies (ie, half of each sample rate). With 384kHz data, the output was down by 16.5dB at 190kHz. Filter 2 was similar to Filter 1, though Filter 3 began to roll off above 16kHz with 44.1kHz data, reaching 3dB at 20kHz (not shwon). Filter 4 had the same premature rolloff above 16kHz with 44.1kHz data (fig.8, green and gray traces), but also rolled off earlier than Filter 1 with the higher sample rates. With 384kHz data, the output was down by 3dB at 85kHz and by 34.5dB at 140kHz.
Channel separation was superb, at >120dB in both directions below 4kHz (not shown). Fed dithered 16-bit data representing a 1kHz tone at 90dBFS via TosLink, the resultant spectrum of the Vega's output showed only the dither noise (fig.9, cyan and magenta traces). Increasing the bit depth to 24 (blue and red traces) dropped the noise floor by an astonishing 28dB, which suggests that the Vega's resolution is commendably close to 21 bits. Some spurious AC-supply components are visible in this graph, particularly in the left channel (blue trace), but as even the highest in level of these spuriae lies well below 130dB, these are inconsequential. When I repeated the test with 24-bit data via the Vega's USB port, I got the same spectrum, confirming that the processor handles 24-bit data via USB. Peculiarly, some very low-level, odd-order harmonics were visible in this spectrum (fig.10) that are absent from fig.9. Although these all lay at or below 146dB, I did wonder if they indicated some problem at the 24th-bit level with USB data. Again, however, these spuriae will be inconsequential.
The spectral peaks at 1kHz in figs.9 & 10 kiss the 90dBFS line, suggesting that the DAC has minimal linearity error. In conjunction with the very low level of analog noise, this allowed the Vega's reproduction of an undithered sinewave at exactly 90.31dBFS (fig.11) to be perfect. The waveform is beautifully symmetrical about the time axis, and the three DC voltage levels defined by these data are superbly well resolved. I was using Filter Mode 4 for this measurement, and you can clearly see this filter's asymmetrical ringing on the waveform tops and bottoms; changing to Filter 1 gave the usual symmetrical Gibbs Phenomenon ringing (fig.12), while increasing the data's bit depth to 24 gave a well-defined sinewave, even at this very low level (fig.13).
Auralic's Vega offered very low levels of harmonic distortion, even when driving a full-scale signal into the very demanding 600 ohm load. Fig.14 reveals that the highest-level harmonic under those circumstances is the third, but that it lies at 116dB, or just 0.0002%! The Vega's performance with the high-frequency intermodulation test depended on the Filter mode used. With Mode 1 (fig.15), not only were the intermodulation products supremely well down in level, with the difference product at 1kHz lying at 128dB (0.00005%), all the ultrasonic images of the 19 and 20kHz tones are all very well suppressed. Mode 2 (not shown) was similar to Mode 1, but with Modes 3 and 4 (fig.16), the ultrasonic images were well evident. Even so, actual intermodulation was as low as it had been in fig.15.
Tested for its rejection of word-clock jitter with the Femto Master Clock set to Auto, the Vega measured superbly well with all inputs, even TosLink. With 16-bit J-Test data (fig.17), all the odd-order harmonics of the low-frequency, LSB-level squarewave were at the correct level, and there were no jitter-related sidebands. With 24-bit datafig.18 shows the output spectrum for data input to the USB portthe spectral spike that represents the 11.025kHz tone is virtually free from the spectral "shoulders" that would suggest the presence of low-frequency random jitter, and any sidebands lie at or below 144dB!
Auralic's Vega D/A processor offers measured performance that is beyond reproach.John Atkinson