Sonic Frontiers SFD-2 D/A processor Mk.I Measurements
I was eager to measure the SFD-2's bench performance, both from curiosity about how the processor would perform technically, and to attempt to measure the jitter-rejection capabilities of the UltraAnalog AES20 input receiver.
The SFD-2's maximum output level when decoding a full-scale, 1kHz sinewave was a high 6.65V from the balanced outputs and 3.32V from the single-ended jacks. This is essentially the D20400A DACs' output levels, since no gain is provided in the tubed output stage. The output-level matching between channels was the best I've measured: the balanced output levels were identical down to the millivolt (6.655V in both channels). The left and right single-ended outputs were within 0.02dB of each other.
I had a hard time measuring the SFD-2's output impedance—it became current-limited and clipped the waveform when driving the Audio Precision's 600 ohm input impedance. We calculate a device's output impedance by measuring the voltage drop when driving a 600 ohm load compared to an open circuit (actually, the AP's maximum 100k ohm input impedance). I then substituted a 2k ohm resistor for the 600 ohm load and measured an output impedance of 310 ohms from the unbalanced outputs at 1kHz and 20kHz. The output impedance rose dramatically, to over 3k ohms at 20Hz. The balanced outputs had an output impedance of 735 ohms at 1kHz and 20kHz, rising to 8k ohms at 20Hz. This is unusual behavior.
To further explore this behavior, I connected a variable-resistance box in parallel with the AP's 100k ohm input and watched the SFD-2's distortion rise as the load impedance was reduced. At 1kHz, 0dBFS, the 0.1% THD value was reached with a load impedance of 2.5k ohms (from a low of 0.006% with any impedance greater than 7.5k ohms) from the balanced outputs. Although these load impedances are far lower than the input impedances of most preamps, they are the values at which the onset of clipping occurs.
A preamplifier with a high input impedance is therefore recommended for the SFD-2. The Audio Research LS2B and LS5 have a balanced input impedance of 3M ohms (three million ohms), making them ideal loads for the SFD-2's high output impedance at low frequencies. When driving low-input-impedance preamplifiers (or, particularly, passive level controls), the SFD-2 may lose dynamics and the bass may get soggy. Going back to my listening impressions comparing the balanced and unbalanced outputs, I must note that while the LS2B's balanced input impedance is 3M ohms, the unbalanced input impedance is 50k ohms. The bass was much better and the dynamics more powerful when listening to the balanced outputs. The LS2B's lower unbalanced input impedance could have contributed to the SFD-2's reduced sonic performance in unbalanced mode.
The SFD-2's excellent channel matching can be seen in fig.1, the SFD-2's frequency response and de-emphasis error; note how the left- and right-channel traces perfectly overlap. The frequency response is unusual in that the response doesn't roll off a few tenths of a dB at 20kHz, as is typical. Interchannel crosstalk, shown in fig.2, was excellent and just within the published specification. I measured a channel separation of 85dB at 10kHz, improving to 105dB at 1kHz—exactly as specified. Both channels have virtually identical crosstalks, which seems to be the result of capacitive coupling between channels (indicated by the smooth 6dB/octave decrease in separation with frequency). Channel separation measured at the single-ended jacks was 7dB lower across the band, measuring 78dB at 10kHz and 98dB at 1kHz.
Fig.1 Sonic Frontiers SFD-2, frequency response (top) and de-emphasis error (bottom) (right channel dashed, 0.5dB/vertical div.).
Fig.2 Sonic Frontiers SFD-2, crosstalk R-L (L-R dashed, 10dB/vertical div.).
Fig.3 is a spectral analysis of the SFD-2's output when decoding a -90dB, dithered 1kHz sinewave. The perfect matching between the DACs is evident by the overlapping traces at the test-signal frequency. The overall noise level is low, but we can see a very low level of 60Hz power-line noise in the right channel, with just a hint of 60Hz noise in the left channel. The right channel also has an unusual little peak at just under 2kHz, though this is insignificant. Performing the same spectral analysis, but with an input signal of all zeros (and, of course, no dither as in the previous trace), produced the plot of fig.4. The slightly higher level of 60Hz power-supply noise is undoubtedly the result of a different grounding scheme between the SFD-2 and the System One. (For this all-zeros test, the processor was driven by a CD player's digital output, not the System One's digital signal generator as in fig.3.) The same tests performed on the single-ended outputs revealed a higher level of 60Hz noise, though this wasn't excessive.
Fig.3 Sonic Frontiers SFD-2, spectrum of dithered 1kHz tone at -90.31dBFS, with noise and spuriae (1/3-octave analysis, right channel dashed).
Fig.4 Sonic Frontiers SFD-2, spectrum of digital silence, with noise and spuriae (1/3-octave analysis, right channel dashed).
The SFD-2's linearity (fig.5) was excellent: the DACs were perfectly linear to -105dB, where noise begins to dominate the measurement. The new UltraAnalog D20400A DAC apparently has linearity as good as its predecessor. Again, this is superb performance. Note that this low-level precision is inherent in the DAC and isn't subject to misadjustment by an MSB trimmer, time, or temperature.
Fig.5 Sonic Frontiers SFD-2, departure from linearity (right channel dashed, 2dB/vertical div.).
The SFD-2's performance on the noise-modulation test was also outstanding (fig.6). The noise level is low, and the traces are very tightly grouped. Note that this measurement was performed using an averaging technique that better reveals the device's noise-modulation performance. This plot indicates that the noise floor doesn't shift as a function of signal level, and that the noise floor's spectral balance remains the same regardless of signal amplitude.
Fig.6 Sonic Frontiers SFD-2, noise modulation, -60 to -100dBFS (10dB/vertical div.).
Unfortunately, the SFD-2 didn't perform quite as well on the 19+20kHz intermodulation test (fig.7). The plot is an FFT-derived spectrum of the SFD-2's output when decoding a full-scale mix of 19kHz and 20kHz sinewaves. The 1kHz difference component (20kHz minus 19kHz) is moderately high in level, and we can see spikes at multiples of 1kHz. Moreover, the level of spurious components between 15kHz and 23kHz is alarmingly high. (For comparison, see p.217 of the October '93 Stereophile, which shows the PS Audio Reference Link's performance on this test.) When measured from the unbalanced outputs, the SFD-2's intermodulation performance degraded further (fig.8), with the 1kHz difference component 15dB higher at -66dB (referenced to full-scale output). (Note, however, that the unbalanced outputs produced fewer lower-frequency IM products. I wondered if this higher level of intermodulation products had anything to do with my impressions of less liquid instrumental textures when listening to the single-ended outputs.
Fig.7 Sonic Frontiers SFD-2, balanced, HF intermodulation spectrum, DC-22kHz, 19+20kHz at 0dBFS (linear frequency scale, 20dB/vertical div.).
Fig.8 Sonic Frontiers SFD-2, unbalanced, HF intermodulation spectrum, DC-22kHz, 19+20kHz at 0dBFS (linear frequency scale, 20dB/vertical div.).
Despite the fact that this is an unrealistic test signal—no musical program would have full-scale energy at 19kHz and 20kHz—the relatively high level of intermodulation products generated by the SFD-2 when confronted by a high-level test signal may correlate with my impression of somewhat harder textures during very loud passages. I did note in my auditioning that the SFD-2 became less liquid when the signal got loud, something I didn't hear from the No.30. Interestingly, the only other digital processor with this high a level of IM products was the tubed VTL Reference converter reviewed in Vol.13 No.12.
The squarewave response, not shown, was typical for the NPC digital filter.
The SFD-2's word-clock jitter—measured using the modified techniques described in last October's Stereophile—varied between 148 and 274 picoseconds, depending on the input signal. The lower figure was measured with a full-scale input signal, the higher with a -90dB input signal. Although the 148ps value is about the same as that measured in the PS Audio Reference Link (which uses the Crystal input receiver), the SFD-2's jitter with a worst-case signal was about half that measured in the Reference Link with the same input signal. Moreover, the SFD-2's jitter spectrum (fig.9, full-scale 1kHz sinewave input) was exceptionally clean, indicating that the jitter is the more sonically benign "white" (random) jitter than the musically degrading periodic jitter (footnote 1).
Fig.9 Sonic Frontiers SFD-2, word-clock jitter spectrum, DC-20kHz, when processing 1kHz sinewave at 0dBFS (linear frequency scale, 10dB/vertical div., 0dB=1ns).)
Characterizing a digital processor's technical performance should include some measurement of how well it rejects jitter from the CD transport. Some processors will pass transport jitter to the recovered clock—the point where it degrades the processor's musical performance—while other processors are better able to reject incoming jitter. The degree to which the digital processor passes incoming jitter may be a good indicator of how sensitive it is to transport differences. With the theoretically perfect input receiver, there should be no sonic differences between transports. With a very poor input receiver, transport differences should be magnified.
We've previously been unable to assess a digital processor's ability to reject transport jitter. Although we've been measuring word-clock jitter inside the processor with the Meitner LIM Detector (see Stereophile, January 1992) for the past year, we had no way of knowing how much jitter was in the S/PDIF or AES/EBU datastream fed to a processor.
Now that we can measure transport jitter (described in Rémy Fourré's "Jitter & the Digital Interface" in October's Stereophile, and in my "A Transport of Delight" last month), we can—for the first time—attempt to quantify how well a digital processor rejects transport and interface jitter.
To assess how well a processor handles incoming jitter, we use the Meitner LIM Detector to measure the digital processor's word-clock jitter when driven by a high-jitter source (the Panasonic SV-3700 DAT machine), and then by a low-jitter source (the PS Audio Lambda CD transport). The SV-3700 has about 4250ps of interface jitter; the Lambda has two orders of magnitude less jitter, measuring about 35ps. Processors with poor transport-jitter rejection should have a larger spread in the measured word-clock jitter between the high- and low-jitter sources. Processors with good transport-jitter rejection should show less difference in their word-clock jitters between high- and low-jitter sources.
Using this hypothesis, I measured the SFD-2's word-clock jitter when driven by the SV-3700, then the Lambda. Fig.10 is the SFD-2's word-clock jitter when driven by a source having 4250ps of interface jitter. The RMS word-clock jitter inside the SFD-2, measured over a bandwidth of 22Hz-30kHz, is 579ps. Fig.11 is the SFD-2's word-clock jitter when driven by the low-jitter Lambda. Although the RMS level dropped to 274ps, the spectrum is virtually identical. The 4kHz component is noticeably lower in level, however.
Fig.10 Sonic Frontiers SFD-2, Panasonic SV-3700 source, word-clock jitter spectrum, DC-20kHz, when processing 1kHz sinewave at -90dBFS (linear frequency scale, 10dB/vertical div., 0dB=1ns).
In short, the jitter spectrum remained virtually the same with these very different levels of jitter in the S/PDIF input signal—an indication of good input-receiver performance. Moreover, the RMS level only doubled when the low-jitter source was replaced with one having jitter two orders of magnitude greater. For comparison, the Reference Link's RMS jitter quadrupled when driven by the high-jitter source. Further, the Reference Link's jitter spectrum was decidedly less clean when driven by the SV-3700. (See pp.218-219 in the October Stereophile for the Reference Link's jitter spectra when driven by the two digital sources.)
Although we don't have nearly large enough a database from which to draw firm conclusions, these measurements seem to indicate that the SFD-2's AES20 input receiver is indeed much better able to reject transport jitter than the Crystal CS8412. The AES20 also produces a clock nearly free from periodic jitter components, regardless of the input signal. Further, the AES20 begins rejecting jitter above 1kHz, which can be seen in figs.10 and 11 as the downward slope of the curve.
Fig.11 Sonic Frontiers SFD-2, PS Lambda source, word-clock jitter spectrum, DC-20kHz, when processing 1kHz sinewave at -90dBFS (linear frequency scale, 10dB/vertical div., 0dB=1ns).
To assess the AES20's jitter attenuation as a function of input jitter frequency, I drove the SFD-2 with input signals of increasing frequency and plotted the word-clock jitter with FFT techniques. With most processors, the jitter spectrum contains a line at the test-signal frequency; the jitter is correlated with the audio signal frequency, resulting in jitter on the word clock with the same frequency as the test signal. With the SFD-2, however, the AES20's very low jitter attenuation cutoff frequency was easily seen. As the test signal rose above 4kHz, the word clock's jitter spectrum showed no trace of correlation with the audio signal. With a 10kHz input signal, there was absolutely no trace of 10kHz jitter component on the SFD-2's word clock.
I also created a new digital test signal for examining the jitter attenuation vs frequency of digital processors. The test signal is made up of low-amplitude (-80dB) digital sinewaves at 500Hz, 1kHz, 2kHz, 4kHz, 8kHz, and 16kHz. By driving the processor under test with this signal and looking for the signal-correlated jitter components in the word clock, we can perhaps gauge how the input receiver attenuates jitter as a function of frequency. Although the SFD-2 seemed to perform well on this test—its clock jitter spectrum was virtually free from signal-correlated jitter above 4kHz—we'll wait to publish the plot until we've performed this test on more digital processors. We may adjust the test-signal frequencies and amplitude to best reveal an input receiver's jitter-rejection performance.
Finally, I measured very low levels of DC at the output: 0.9mV and 1.5mV from the left and right balanced outputs, respectively, and 0.6mV from both left and right unbalanced outputs. The SFD-2 had no problem locking to 32kHz and 48kHz sampling rates. As mentioned earlier, the polarity-inversion switch in my sample was either miswired or installed upside-down; the 0-degree front-panel position produced an inverted output signal from the unbalanced jacks. The SFD-2's output XLR jacks are wired according to the AES/ANSI standard with pin 2 "hot."
Overall, the SFD-2's bench performance was excellent. I was concerned, however, about the SFD-2's very high output impedance at low frequencies, which suggested that the SFD-2 will work its best when driving preamplifiers with high input impedance (the ARC LS5's 3M ohm impedance was ideal). With preamplifiers having much lower input impedances, the Sonic Frontiers' balance could become a little lean. Further, the relatively high level of intermodulation products—particularly from the unbalanced outputs—is far from ideal behavior.—Robert Harley
Footnote 1: Ed Meitner has nearly finished a version 3 LIM Detector that has intrinsic jitter an order of magnitude lower than the current instrument. This should help in increasing the measuring accuracy of very-low-jitter clocks. The new LIM Detector also has a bandwidth of 40kHz, allowing us to look at the entire sonically significant jitter spectrum.—Robert Harley