MSB Technology Platinum Data CD IV transport & Diamond DAC IV & D/A converter Measurements
Sidebar 3: 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 measure the MSB Diamond DAC IV; for some tests, I also used my vintage Audio Precision System One Dual Domain.
The MSB combination offers a bewildering number of playback options, with three different flavors of digital data input (USB, AES/EBU and S/PDIF, and MSB Network), four reconstruction filters, two choices of upsampling, and both balanced and unbalanced outputs. I started my testing, therefore, with the more straightforward balanced analog input. This offered a usefully high balanced input impedance of 212k ohms and preserved absolute polarity with the Diamond DAC IV set to Non-Invert. The maximum gain was 9.1dB with the volume control set to "9dB," with "0dB" giving unity gain, as expected; the volume control operated in accurate 1dB steps, with superb matching between channels. The frequency response was perfectly flat up to the 200kHz limit of the Audio Precision generator, and channel separation was superb, at >118dB LR and >132dB RL in the audioband. The Diamond DAC IV's output impedance was 50 ohms at all audio frequencies from the RCA jacks and 100 ohms from the XLRs, the latter wired with pin 2 hot.
The Platinum Data CD IV disc transport offered excellent error correction, there being no glitches in its output until the gaps in the data spiral of the Pierre Verany Records Test CD reached 2.4mm in length. I didn't have issues playing audio CDs. However, while the transport correctly recognized the sample rate and bit depth of WAV files when I played a selection of test files I'd burned to a DVD-R, a data CD-R with the same files was played at a constant 44.1kHz with the wrong frequency; eg, a 1kHz tone sampled at 176.4kHz was reproduced as a 250Hz tone sampled at 44.1kHz.
Returning to the Diamond DAC IV, the Macintosh USB Prober utility reported that it did operate in isochronous asynchronous mode, as advertised. The manufacturer string was identified as "MSB," the product string as "MSB USB Audio 2.0." AudioMIDI Set-Up indicated a maximum sample rate of 384kHz! The output level with a 0dBFS digital signal and the volume control set to "0dB" was 5.29V from the balanced jacks, 2.64V from the unbalanced. (The latter was correctly set with the Set-Up Menu, so that it provided the sum of the balanced output stages rather than just the hot phase.) Both outputs preserved absolute polarity unless the polarity was inverted with the appropriate function button on the remote. Channel separation (not shown) was superb, at >125dB in both directions below 1kHz, and still 114dB at 20kHz. (The volume control was set to "0dB" for all tests; when I refer to upsampling, this was option 2, labeled U2.)
Fig.1 shows the impulse response of the first reconstruction filter, taken with 44.1kHz data and no upsampling. Filter 2 was very similar, both filters offering a conventional time-symmetrical impulse. Filter 3, labeled Lanczos 3, had a single cycle before and after the main impulse (fig.2), which is typical of a slow-rolloff type; Filter 4, labeled Minimum-Phase, had, as expected, a minimum-phase impulse response (fig.3) and no pre-ringing.
Filters 1, 2, and 4 had very similar frequency responses. Fig.4, for example, shows the response with data sampled at 44.1, 96, and 192kHz, with no upsampling. With each sample rate, the output is flat within the audioband but with then an early, very steep rolloff just below the Nyquist frequency (half the sample rate). I assume this is due to the apodizing nature of the filters, which will have a complete null at Nyquist. This does mean that with 44.1kHz data, while Filter 2's response extends at full level to 20kHz, with Filters 1 and 4 the output is full-level up to 17.5kHz then down by 12dB at 20kHz. To my surprise, switching in Upsampling 2 restricted the bandwidth to 16kHz (not shown; Upsampling 1 behaved correctly). Fig.5 shows the frequency response at the same three sample rates with Filter 3 and no upsampling. There is now a very slight rise before the rolloff at each rate, and the rate of rolloff, as expected from the impulse response, is slow. The peculiar stepped shape of the rolloff is, I suspect, due to the filter's "leaky" nature allowing aliasing energy to affect the measurement.
For reasons of consistency with my measurements of digital products going back to 1989, my first examination of a product's resolution is to analyze the analog output by sweeping a 1/3-octave bandpass filter from 20kHz to 20Hz while it reconstructs a dithered 1kHz tone at 90dBFS with 16- and 24-bit data. With 16-bit data (fig.6, top pair of traces), the MSB's spectrum peaked at exactly 90dBFS, suggesting minimal linearity error, this confirmed by a separate test (not shown). All the traces are showing is the level of the 16-bit dither noise; increasing the bit depth to 24 drops the noise floor by an extraordinary 27dB or so. This implies a resolution of close to 21 bits, which is not only enough to allow the Diamond DAC IV to correctly resolve a 24-bit/1kHz tone at 120dBFS (fig.6, bottom traces), but is one of the highest resolutions I have encountered. (Only the NAD M51, reviewed in the July 2012 issue, and the Weiss DAC202, reviewed last January, were as good as the MSB.)
However, the increase in bit depth has unmasked low-level power-supply spuriae at 60 and 180Hz. These were not affected by floating the analog ground with the Set-Up Menu; I suspect they are due to magnetic interference from the transformers in the Diamond Power Base, which sat beneath the DAC IV for these tests. Repeating the spectral analysis with a modern FFT technique (fig.7) confirmed both the MSB's extraordinary resolution and the presence of the low-level supply components, though it's fair to note that, at below 130dB, these are of only academic interest. I was a little surprised by the presence of some second- and fourth-harmonic content with the 24-bit data in fig.7 (blue and red traces). Switching in upsampling increased the level of these harmonics by 5dB or so (not shown), but again, this is of academic interest.
The DAC IV's reproduction of an undithered tone at exactly 90.31dBFS with 16-bit data was essentially perfect (fig.8), the symmetrical waveform clearly resolving both the three DC voltage levels described by the data, and Filter 1's Gibbs Phenomenon ringing at the transition points. With undithered, 24-bit data, the MSB output a superbly clean, low-noise sinewave (fig.9), despite the very low signal level and the lack of dither.
With high-level signals, the Diamond DAC IV produced more harmonics than I was expecting (fig.10). Yes, the harmonic highest in level was the third, at a very low 91dB (0.003%), and the measured THD+noise percentage was just 0.005%, but I don't like to see those higher-order components. This graph was taken with Upsampling 2 enabled; the picture didn't change without upsampling or at higher frequencies.
The MSB's performance on the high-frequency intermodulation test, in which the device under test decodes a full-scale, 24-bit mix of 19 and 20kHz tones, depended on both the reconstruction filter in use and on upsampling. Fig.11, for example, shows the spectrum of the Diamond DAC IV's output with no upsampling and Filter 3. The filter's slow rolloff has reduced the levels of the fundamental tones, while the ultrasonic images of the tones lie at almost the same levels. In addition, while actual intermodulation products are low in level, the audioband is littered with aliasing products. For comparison, fig.12 shows the spectrum with Filter 4, again without upsampling. While there are now no aliasing products in the audioband, the spectrum is not as clean as I would have liked to see. Switching in Upsampling 2 (fig.13) drops the level of the 20kHz tone as expected from the response tests, and the spectrum is cleaner overall.
The Diamond DAC IV demonstrated excellent rejection of word-clock jitter via all of its digital inputs. I found no appreciable differences among the AES/EBU, USB, and MSB Network presentations of the J-Test data. However, while there was no accentuation of data-related spuriae, and the single spectral spike representing the Fs/4 tone was superbly well defined, there were persistent sidebands, low in level but of unknown origin at ±1222 and ±1446Hz (fig.14), regardless of the input used.
In most ways, MSB Technology's Diamond DAC IV offers the best measured performance in the digital domain that I have encountered.John Atkinson