dCS Scarlatti SACD/CD playback system Measurements
I assessed the performance of the dCS Scarlatti system using Audio Precision's SYS2722 system (see www.ap.com and "As We See It" in the January 2008 issue), as well as, for some tests, our Audio Precision System One Dual Domain.
Because of the complexity of this four-box system and the multitude of operational choicesinternal or external clock (the latter with or without dither), SACD or CD playback or external 16-bit and 24-bit data via S/PDIF, AES/EBU, or USB, with PCM data upsampled to 88.2kHz, 96kHz, 176.4kHz, or 192kHz PCM or to or DSD, with four different filters offered by both the DAC and the upsampler, and both balanced and unbalanced analog outputs, each with a choice of two maximum output levelsI had to narrow down the choices for the measurements. For evaluating the system's performance with SACD, I used the Sony "provisional" test SACD, with both the Scarlatti transport and Scarlatti DAC clocked by the Scarlatti Clock. For CD playback, I played a test CD-R in the transport, either fed directly to the DAC or upsampled by the Scarlatti Upsampler. I also fed 16-bit and 24-bit PCM data to the Upsampler via a Toslink connection from a PC with an RME soundcard. Again, I either upsampled these data to both 176.4kHz and DSD, or fed them directly to the Scarlatti DAC. I also drove the Scarlatti Clock's USB port with a MacBook running OS10.4.11, using both 44.1kHz and 88.2kHz files played from iTunes 8. I primarily tested the Scarlatti DAC from its balanced outputs with the volume control set to its maximum, but I also repeated a subset of the tests from the unbalanced jacks. Phew!
The Scarlatti transport featured the best error correction I have encountered. It played back the Pierre Verany test CD, which has laser-cut gaps in the data spiral of varying lengths, without any glitches or mutes through Track 36, which has 2.5mm gaps. It did mute once per disc revolution on Track 37, which features 3mm gaps. This transport should be able to play even the most beat-up CDs in your collection without missing any music.
The Scarlatti DAC's maximum output level at 1kHz was 1.975V or 6.07V from both balanced and unbalanced outputs, depending on whether the output level was set to "2V" or "6V." Both outputs preserved absolute polarity, ie, were non-inverting. (The XLRs are wired with pin 2 hot.) The balanced output impedance was very low, at less than 2 ohms. While the unbalanced output impedance was higher, at 51 ohms across the band, this is still low in absolute terms.
The Scarlatti DAC's frequency response with SACD data (fig.1, top pair of traces above 40kHz), was flat almost to 40kHz, with then a gentle rolloff reaching 3dB at 70kHz and 26dB at 107kHz. This was the same with all four reconstruction filters. With CD data upsampled to DSD, the response was flat from 10Hz to 20kHz, as was its response with a pre-emphasized disc, so I haven't shown these in fig.1. With non-upsampled playback, there were very minor changes at the top of the audioband, depending on which of the four filters was chosen. However, there were considerable differences in behavior above the audioband when the Scarlatti DAC was fed non-oversampled 96kHz data. F1 (fig.1, second from top pair of traces at 40kHz) gave the most extended ultrasonic output, before beginning a very steep rolloff above 42kHz. The other three filters all gave a slight increase in the audioband level, with F2 peaking at 42kHz before rolling off very steeply, F3 having the most restricted ultrasonic response, reaching 3dB at 27kHz, and F4 offering more extension than F3 but a slower rate of rolloff than F1. Channel separation was superb, at >120dB below 1kHz, with the RL leakage being even lower than the LR.
Fig.1 dCS Scarlatti, frequency response at 3dBFS into 100k ohms, SACD data (top above 50kHz), and 96kHz PCM data with (from top to bottom at 40kHz): Filter 2, Filter 1, Filter 4, and Filter 3. (Right channel dashed; 2dB/vertical div.)
For consistency with my library of digital previews going back to 1989, my first test of a component's ultimate resolution is to sweep a 1/3-octave bandpass filter from 20kHz to 20Hz while the player decodes dithered data representing a 1kHz tone at 90dBFS. The top pair of traces in fig.2 shows the result with non-oversampled CD data. The traces peak at exactly 90dBFS, suggesting very low linearity error, and are totally free from harmonic or power supply-related spuriae. In fact, all these traces actually show is the effect of the recorded dither. Increasing the bit depth to 24 gives the middle pair of traces in this graph. The noisefloor has dropped by 18dB, implying resolution close to 19 bits, which is superb, and the Scarlatti DAC has no problem decoding a tone at 120dBFS (bottom pair of traces). Again the traces are free from spuriae of any kind. With SACD playback (top traces above 5kHz), there appears to be less resolution than with CDs, but this is a measurement artifact, the format's massive noiseshaping leading to ultrasonic noise leaking beneath the bandpass filter's "skirts." However, there is a little more LF noise apparent with SACD playback.
Fig.2 dCS Scarlatti, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit CD data (top), SACD data (middle at 2kHz), and 24-bit data (bottom), with dithered 1kHz tone at 120dBFS with 24-bit data (bottom at 1kHz). (Right channel dashed.)
Fig.3 was generated using 16- and 24-bit versions of the 1kHz tone at 90dBFS, but this time with the PCM data upsampled to DSD and analyzed using an FFT technique. The reduction in noise with the increase in word length can again be seen, though this is not quite as dramatic as that seen in fig.2, due to the output level being set to 2V rather than the 6V used for fig.2. However, while no harmonic or supply-related spuriae can be seen, there is now a discrete tone apparent at approximately 625Hz with both word lengths. This tone lies at 118dBFS, so it might be thought harmless, but it shouldn't be there at all. Fig.4 repeats the analysis with SACD-derived data, using tones at 90dBFS and 120dBFS. Again, the spurious tone can be seen with both signals.
Fig.3 dCS Scarlatti, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (left channel cyan, right channel magenta) and 24-bit data (left blue, right red), both upsampled to DSD.
Fig.4 dCS Scarlatti, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS (red) and at 120dBFS, DSD data (blue).
I suspect that this tone is some kind of idle pattern occurring in the DSD conversion, as it is absent from an FFT analysis of the16- and 24-bit data used to generate fig.3 now being upsampled to 176.4kHz rather than DSD and fed to the DAC from the Scarlatti Upsampler via a dual-AES/EBU link rather than Firewire (fig.5). (This graph was taken with the DAC set to the 6V maximum output, which gives the same 18dB drop in the noisefloor seen in fig.2.)
Fig.5 dCS Scarlatti, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (left channel cyan, right channel magenta) and 24-bit data (left blue, right red), both upsampled to 176.4kHz PCM.
Testing the Scarlatti DAC's linearity error with 16-bit data only showed the effect of the recorded dither, so I haven't shown it. With 24-bit data (also not shown), the amplitude error was negligible down to 120dBFS. Playing back an undithered 16-bit tone at exactly 90.31dBFS (fig.6), the three DC voltage levels are clearly resolved, with excellent waveform symmetry, though a very slight degree of DC offset (40µV right, 10µV left) can be seen. With SACD data (fig.7), there is a good facsimile of the sinewave.
Fig.6 dCS Scarlatti, waveform of undithered 1kHz sinewave at 90.31dBFS, CD data (left channel blue, right red).
Fig.7 dCS Scarlatti, waveform of dithered 1kHz sinewave at 90dBFS, SACD data (left channel blue, right red).
The Scarlatti DAC's analog output stage offers very low distortion, with only the second and third harmonics apparent above 120dB (0.0001%), even into the very low 600 ohm load (fig.8). Though the third harmonic is the highest in level, at 112dB (0.00025%), this is going to be well below audibility. Neither the low level of harmonic distortion nor its spectrum was affected either by the upsampling or by the filter choices.
Fig.8 dCS Scarlatti, balanced output, spectrum of 50Hz sinewave at 0dBFS into 200k ohms, 24-bit data (left channel blue, right red; linear frequency scale).
However, there were some changes in behavior with the HF intermodulation test results. Fig.9 shows the spectrum of the DAC's output while it decoded a maximum-level 24-bit mix of 19kHz and 20kHz tones, fed directly to the DAC without upsampling from the original 44.1kHz and with the Filter set to F1. Intermodulation products are extremely low, lying close to 120dB, and there are no images of the fundamental tones apparent between 21kHz and 24kHz. Fig.10 shows the spectrum with the same data upsampled to DSD. Intermodulation products remain miniscule, but a slight shaping of the noisefloor can be seen above 20kHz. It can't be seen at the scale these graphs are printed in the magazine, but the peaks representing the 19 and 20kHz tones have slightly less wide "skirts" in fig.10 than in fig.9, which suggests that the DSD-upsampled data has even lower jitter. Upsampling the data to 176.4kHz PCM rather than DSD eliminates the noise-floor shaping (fig.11), but the spectral peaks now have slightly wider "skirts."
Fig.9 dCS Scarlatti, balanced output, Filter 1, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms, 24-bit data (left channel blue, right red; linear frequency scale).
Fig.10 dCS Scarlatti, balanced output, Filter 1, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms, 24-bit data upsampled to DSD (left channel blue, right red; linear frequency scale).
Fig.11 dCS Scarlatti, balanced output, Filter 1, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms, 24-bit data upsampled to 176.4kHz PCM (left channel blue, right red; linear frequency scale).
The Filter setting had no effect on the results of the HF intermodulation spectral analysis when the data were upsampled to DSD or 176.4kHz PCM. However, whereas F1 offered textbook performance with non-oversampled 24-bit data (fig.9), the other three filters introduced major spectral changes. Fig.12, for example, shows the behavior with F4, which, with its slow rolloff, gives the lowest rejection of ultrasonic image energy and the greatest degree of "leakage" into the audioband. Many aliasing products can be seen below 20kHz; while this test is very much a worst-case situation, and unlikely to be encountered with music, it is instructive to see the tradeoff this filter requires for better time-domain performance. F3 is better in this respect than F4 and F2 is better still, though neither approaches the spectral purity offered by F1.
Fig.12 dCS Scarlatti, balanced output, Filter 4, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms, 24-bit data, no upsampling (left channel blue, right red; linear frequency scale).
I examined the Scarlatti DAC's rejection of word-clock jitter in a number of circumstances. Fig.13 was generated for the best case situation: CD playback in the Scarlatti transport, upsampled to DSD, with Transport, Upsampler, and DAC all controlled by and locked to the Scarlatti Clock. All the harmonics of the LF squarewave lie at the residual level and the central spike in this graph that represents the high-level, 16-bit 11.025kHz tone is very sharply defined. By contrast, fig.14 shows the worst case, with the 16-bit analytical signal fed to the DAC via 15' of Toslink from a PC fitted with an RME soundcard. The central peak now has noticeably widened "skirts," due to the presence of LF random jitter, sidebands can be seen at ±60Hz, and there is some modulation of the lowest-order sidebands at ±229Hz. In absolute terms, however, this is still very good performance. Feeding the Toslink data to the Upsampler rather than to the DAC and upsampling to DSD eliminates the sidebands (fig.15) and gives a spectrum closer to that shown in fig.13.
Fig.13 dCS Scarlatti, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit CD data upsampled to DSD, master Scarlatti Clock. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.14 dCS Scarlatti, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit data sourced via Toslink, no upsampling. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.15 dCS Scarlatti, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit data sourced via Toslink, upsampled to DSD. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Finally, I repeated most the tests feeding 16- and 24-bit data, sampled at 44.1kHz or 88.2kHz from my MacBook to the Upsampler's USB input. This operates in the much-preferred asynchronous mode, where the DAC controls the flow of data from the computer. In all respects, there were no measured differences between the Scarlatti DAC accepting data from the MacBook via USB and playing the same data on a CD with the Transport or feeding it to an AES/EBU or S/PDIF input. And the efficacy of the asynchronous USB protocol can be seen in fig.16, where the spectrum is almost as clean as that in fig.13.
Fig.16 dCS Scarlatti, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit data sourced from MacBook via USB. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
The dCS Scarlatti may be the most expensive disc-playing system Stereophile has reviewed by far, but it offers state-of-the-art measured performance.John Atkinson