Sonic Frontiers SFCD-1 CD player Measurements
The SFCD-1's output voltage was a rather low 1.9V from the single-ended outputs and 3.8V from the balanced jacks. The left- and right-channel output voltages were matched to 0.02dB, which is essentially perfect.
Unlike the SFD-2 and SFD-2 Mk.II, the Sonic Frontiers CD player had a moderate output impedance of 165 ohms (unbalanced) and 390 ohms (balanced) at any audio frequency. The original SFD-2's output impedance was 735 ohms over most of the band, but rose to a ridiculously high 8k ohms at 20Hz. This high output impedance in the bass was due to the 3µF coupling capacitor. The Mk.II version replaced the 3µF cap with one of 10µF, which along with other changes, reduced the output impedance to 365 ohms over most of the band, rising to 1350 ohms at 20Hz. This is still a high output impedance, but a far cry from 8k ohms. The new direct-coupled topology in the SFCD-1 dramatically decreased the player's source impedance—a good move in my view.
The SFCD-1 doesn't invert absolute polarity (and there's no polarity inversion control), and DC levels were negligible. On the Pierre Verany Test CD, the SFCD-1 could play track 32 without skipping, but faltered on track 33. This is only moderate tracking performance.
Fig.1 shows the SFCD-1's frequency response and de-emphasis response. The 0.6dB rolloff at 20kHz is a little steeper than many processors and is greater than the 0.3dB specified. The de-emphasis circuit produced tracking errors of 0.7dB at 4kHz, with a shelved-down treble region. This will definitely be audible as a reduction in brightness when CDs recorded with pre-emphasis are played. Note that de-emphasis is performed in the digital domain by the Philips chipset that comes with the CDM 12.4 transport mechanism, and can't be turned off if the designer wants to implement analog-domain attenuation. The de-emphasis issue has, however, become a moot point; very few CDs are pre-emphasized.
Fig.1 Sonic Frontiers SFCD-1, frequency response (top) and de-emphasis response (bottom) (right channel dashed, 0.5dB/vertical div.).
The player's channel separation (fig.2) was excellent, measuring 120dB at 1kHz, and better than 106dB at 16kHz. The SFCD-1 does a pretty good job of keeping power supply noise out of the audio circuits, as seen in fig.3, a 1/3-octave spectral analysis of the SFCD-1's output when processing a 1kHz –90dB dithered sinewave. We can see some noise at 60Hz, 180Hz, and 300Hz (particularly in the right channel), but these noise components are low in level. With a test signal of all zeros and a wider-bandwidth analysis (fig.4), we can better see the power-supply noise. Without the dither noise of the 1kHz,–90dB test signal, the noisefloor appears lower, highlighting any power-supply components.
Fig.2 Sonic Frontiers SFCD-1, crosstalk (R-L channel dashed, 10dB/vertical div.).
Fig.3 Sonic Frontiers SFCD-1, spectrum of dithered 1kHz tone at –90.31dBFS, with noise and spuriae (16-bit data, 1/3-octave analysis, right channel dashed).
Fig.4 Sonic Frontiers SFCD-1, spectrum of digital silence (16-bit data, 1/3-octave analysis, right channel dashed).
The player's linearity (fig.5) was basically excellent, though with some slight negative error below –80dBFS. (This can also be seen in fig.3.) Only the left channel is shown; the right channel was identical. The SFCD-1 also produced a superb-looking waveform when processing a 1kHz –90dB undithered sinewave (fig.6). Note the low noise and nearly perfect symmetry between the positive and negative-going phases of the waveform.
Fig.5 Sonic Frontiers SFCD-1, left channel, departure from linearity (2dB/vertical div.).
Fig.6 Sonic Frontiers SFCD-1, waveform of undithered 1kHz sinewave at –90.31dBFS (16-bit data).
When decoding data representing a full-scale mix of 19kHz and 20kHz tones, the SFCD-1 produced the intermodulation distortion spectrum of fig.7. Although the 1kHz component reaches the –90dB level, the rest of the spectrum is clean.
Fig.7 Sonic Frontiers SFCD-1, HF intermodulation spectrum, DC–24kHz, 19+20kHz at 0dBFS (linear frequency scale, 20dB/vertical div.).
I was particularly interested in the SFCD-1's measured jitter, both because of Sonic Frontiers' claims for the unit, and as a test of the correlation between the Meitner LIM Detector and UltraAnalog's jitter analyzer at low jitter levels.
Fig.8 is the SFCD-1's jitter spectrum when decoding a 1kHz full-scale sinewave, measured with the Meitner instrument. As you can see, the spectrum is completely free from periodic jitter components, which would appear as spikes in the trace. The RMS jitter level, measured over a 400Hz–20kHz bandwidth, was just 17 picoseconds.
Fig.8 Sonic Frontiers SFCD-1, word-clock jitter spectrum, DC–20kHz, when processing 1kHz sinewave at 0dBFS; (linear frequency scale, 10dB/vertical div., 0dB=1ns).
Varying the test signal had no effect on the jitter's spectrum or RMS level. Figs.9 & 10 show the jitter spectrum when the SFCD-1 was decoding digital silence and a 1kHz –90dB sinewave respectively. Note that the spectra look identical, and that the RMS jitter level remained at 17ps regardless of the signal being processed. This low RMS jitter level, along with the absence of periodic components with any test signal, are unique in my experience.
Fig.9 Sonic Frontiers SFCD-1, word-clock jitter spectrum, DC–20kHz, when processing digital silence; (linear frequency scale, 10dB/vertical div., 0dB=1ns).
Fig.10 Sonic Frontiers SFCD-1, word-clock jitter spectrum, DC–20kHz, when processing 1kHz sinewave at –90dBFS; (linear frequency scale, 10dB/vertical div., 0dB=1ns).
The SFCD-1's jitter spectra as measured by the UltraAnalog analyzer show some signal-correlated periodic components not seen in the LIM Detector spectra. Compare figs.8 & 10 (made with the LIM Detector) with figs.11 & 12 (made with the UltraAnalog analyzer). The low-level periodic components seen with the latter, but not in the former, are revealed because of the UltraAnalog analyzer's lower intrinsic jitter (1.2ps). The spikes may exist in fig.8 & 10 but lie beneath the LIM Detector's noise floor.
Fig.11 Sonic Frontiers SFCD-1, word-clock jitter spectrum, DC–20kHz, when processing 1kHz sinewave at 0dBFS; measured with UltraAnalog Jitter B analyzer (linear frequency scale, 1mV/ps).
Fig.12 Sonic Frontiers SFCD-1, word-clock jitter spectrum, DC–20kHz, when processing 1kHz sinewave at –90dBFS; measured with UltraAnalog Jitter B analyzer (linear frequency scale, 1mV/ps).
The RMS jitter level as measured with the UltraAnalog jitter analyzer was 8ps. The discrepancy between this figure and the 17ps I measured with the LIM Detector is understandable. At these low jitter levels, accurate readings are difficult with the LIM Detector: a few tenths of a millivolt in the measured reading can double the RMS jitter calculation when the jitter level is so low.
The input to the LIM Detector is a 10x probe attached to the 8x (usually) word clock driving the DAC. With UltraAnalog DACs, the crucial timing reference is called the "de-glitch" clock. The LIM Detector output is fed into one of the Audio Precision System One's analog inputs. The System One's bandlimiting filters are invoked at 400Hz and 22kHz—400Hz to remove some noise in the LIM Detector, and 22kHz because that is the LIM Detector's intrinsic bandwidth. The RMS voltage output from the LIM Detector is read from the computer monitor connected to the System One, then converted into the RMS jitter level using a conversion factor of 311mV = 1 nanosecond of jitter. The calculated number, however, represents the jitter of the device under test combined with the LIM Detector's intrinsic jitter.
When two uncorrelated noise sources are added, the total noise energy is equal to the square root of the sum of the squares of the individual noise sources. By squaring the measured RMS jitter level, subtracting the square of the LIM Detector's intrinsic jitter, then taking the square root of the result, we arrive at the RMS jitter level of the device under test with the LIM Detector's intrinsic jitter removed. As you can imagine, even tiny changes in the measured voltage make large differences in the resultant figure when the jitter levels are near the LIM Detector's noise floor. For example, the difference between UltraAnalog's 8ps figure and the LIM Detector's 17ps figure is about 500µV, a difficult distinction to resolve when the readout has some bounce. With measurements on higher-jitter products, the final result is less sensitive to small voltage variations.
With its intrinsic jitter of just 1.2ps, UltraAnalog's jitter analyzer doesn't need such a calculation to arrive at the RMS jitter level. Instead, its output is calibrated so that 1mV = 1ps of jitter.
At any rate, the SFCD-1's jitter was by far the lowest I've measured, its jitter spectrum was the cleanest I've seen, and the clock is nearly immune to being jittered by signal-correlated effects.
Overall, the SFCD-1 performed well on the bench in aspects such as crosstalk, linearity, and reproduction of low-level signals, although the spectrum analysis showed a trace of power supply noise in the audio signal.—Robert Harley