Sidebar 5: Measurements
For lower 44.1kHz–192kHz rates, F1-F4 are linear phase for all sample rates and offer responses of –22dB/45kHz (but flat to 40kHz), –0.6dB/45kHz, –12dB/34kHz and –10.5dB/45kHz, respectively, with 96kHz media, and –13dB/90kHz, –7.0dB/90kHz, –14dB/68kHz and –17.6dB/90kHz (fig.4, black, red, cyan and green traces), respectively, with 192kHz files. F4 is a low-tap linear phase filter that trades reduced time-domain distortion for worse stopband rejection (red spectrum, fig.3). F5 is the outlier here, a fast minimum-phase filter (orange traces, fig.4) that rolls away early to –1.5dB/20kHz and –14dB/60kHz with 192kHz media while simultaneously offering superb stopband rejection of 125dB (black spectrum, fig.3).
Fig.1 Distortion + noise versus 24-bit/48kHz digital signal level over a 120dB range (black, 1kHz; blue, 20kHz). Note 140dB Y axis.
With the separation of the Varèse's core components—not least the evolution of the Ring DAC into two mono, differential arrays—we would expect to witness some uplift in technical performance. And indeed, comparisons with the Vivaldi APEX DAC show incremental gains in S/N ratio, distortion, and low-level linearity. Specifically, while the 5.88V maximum output (in the Varèse's top 6V mode) has not changed, and the sub–1 ohm balanced source impedance is retained, second-harmonic distortion through the bass and midrange has fallen from 0.00004% to 0.00003% over the top 10dB of the Varèse's dynamic range (see fig.1). Such figures are of academic interest only of course. The 20kHz figure remains unchanged at 0.0001%, as this is a function of the final analog output buffer and not a digital nonlinearity. However, the Varèse balanced Ring DAC has delivered a 1dB uplift in A-weighted S/N ratio from 117.1dB to 118.1dB, while low-level resolution is now good to ±0.2dB at –120dBFS. This is all reflective of a state-of-the-art DAC.
Fig.2 High-resolution jitter spectrum (24-bit/48kHz data). Very minor PSU-related sidebands and a pair at ±1.95kHz only.
The Vivaldi APEX's elimination of random jitter/phase noise exposed an increase in correlated jitter of 120ps (24-bit/48kHz data), so it's gratifying to report that the Varèse is having its cake and eating it, too: Phase noise remains almost entirely absent, while the Vivaldi's ±5Hz, ±7Hz, ±9Hz, etc, sidebands are now fully suppressed. A mere ~5ps of jitter remains at ±1.95kHz and ±120Hz (a PSU rectifier component) over all sample rates (see fig.2). Clock accuracy and digital jitter are not necessarily linked, but in this case, the Varèse's Master Clock system holds true to within ±1ppm. Incidentally, Class-1 accuracy is specified within ±50ppm!
Fig.3 Stopband image at 51kHz in response to 45kHz/24-bit signal sampled at 96kHz. Rejection is 125dB (Filter F5, black) vs. 5.2dB (Filter F4, red). Note the second harmonic at 90kHz and the image at 96kHz.
Otherwise, the Varèse's frequency response(s), stopband rejection, and time domain behavior all depend on your choice of dCS's six adaptive digital filters, F1 to F6, which are carried over from the latest version of the Vivaldi. The filter coefficients change with some sample rates, but as a rule, the linear phase F2 filter offers the flattest responses, out to ±0.00dB/20kHz, –0.6dB/45kHz and –7.0dB/90kHz with 48kHz, 96kHz and 192kHz media, respectively, this at the expense of the poorest image rejection among the filters (just 3.7dB with 44.1kHz/48kHz sample rates).
Fig.4 Time (impulse) and frequency responses with 192kHz data. Filters F1 (black); F2 (red); F3 (cyan); F4 (green); F5 (orange); and F6 (purple).
F6 is a fast apodizing/linear phase type (purple traces, fig.4) with very extended ringing but full, >125dB stopband rejection and responses that match F5 at ±0.02dB/20kHz, and –3.9dB/45kHz (flat to 43kHz) with 48kHz and 96kHz files, respectively, and F3 at 192kHz, where they both roll away to –14dB/60kHz.
Importantly, all these filters passed the intersample clipping test, suggesting that dCS has built at least 3dB of digital headroom into the latest Ring DAC.—Paul Miller
