MSB Platinum Link Plus D/A processor Measurements part 2

Harmonic distortion was very low, even into the punishing 600 ohm test load (fig.7), where the highest-level harmonic was the second, at -94dB (0.002%). However, I was a little bothered by the appearance of higher-order harmonics in the spectrum, so I repeated the test using a higher frequency and an analyzer with a higher dynamic range than the Audio Precision System One. Fig.8 shows that the MSB does indeed produce some odd-order harmonics. While these are still at a very low level, they will, all things being equal, be more audible than the more usual second and third harmonics, due to the reduced level of masking by the primary signal. Intermodulation distortion was very low (fig.9), if not quite to the vanishingly small levels of the Weiss Medea.

Fig.7 MSB Platinum Link Plus, balanced, spectrum of 50Hz sinewave, DC-1kHz, at 0dBFS into 600 ohms (linear frequency scale).

Fig.8 MSB Platinum Link Plus, unbalanced, spectrum of 1kHz sinewave, DC-10kHz, at 0dBFS into 8k ohms (linear frequency scale).

Fig.9 MSB Platinum Link Plus, balanced, HF intermodulation spectrum, DC-24kHz, 19+20kHz at 0dBFS into 600 ohms (linear frequency scale).

I used the Miller Audio Research Jitter Analyzer to examine the MSB Platinum's rejection of word-clock jitter. The result was very dependent on the datalink used. Feeding the Platinum from my reference PS Audio Lambda CD transport via a 6' coaxial datalink gave a low 197 picoseconds of peak-peak jitter. The spectrum of this jitter is shown in fig.10; it is mainly data-related (red numeric markers), though sidebands at ±15.6Hz (purple "1") and ±120Hz (blue "4") contribute to the measured figure.

Fig.10 MSB Platinum Link Plus, unbalanced, high-resolution jitter spectrum of analog output signal, PS Audio Lambda via 6' S/PDIF datalink (11.025kHz at -6dBFS sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. (Grayed-out trace is with WAV-file playback on PC fitted with RME Digi96/8 PRO soundcard via 15' plastic TosLink.)

But when I changed the data source to WAV-file playback on my PC, using an RME Digi96/8 PRO soundcard to drive the MSB via a 15' length of plastic TosLink, the word-clock jitter shot up to 572ps. Not only did the primary data-related sidebands at ±229Hz rise alarmingly—shown by the grayed-out trace in fig.10—but a slew of low-frequency sidebands made an appearance around the central tone. The moral is obvious: Owners of the MSB Platinum need to use a good coaxial datalink and steer clear of the bandwidth-limited TosLink connection.

While I was performing the jitter measurements, I happened to glance at the screen of the analog oscilloscope that I use to monitor the output of the device under test and noticed something unusual. Fig.11 shows both the data points and the ideal reproduced waveform of the analytical signal used by the Miller Analyzer. Developed by the English engineer Julian Dunn, the signal consists of a high-level tone at exactly one quarter the sample frequency, Fs/4, which is therefore represented by the same four data points in each waveform cycle. (This is overlaid by a 229.68Hz squarewave, Fs/192, at the LSB level, which maximally tests a DAC's rejection of word-clock jitter.) It is the interaction between these sparse data points and the DAC's low-pass filter (see my article on the subject in our online archives) that completely reconstructs the analog waveform. That this is the case is demonstrated by fig.12, which shows the reconstructed output waveform from a conventional D/A processor: a perfect sinewave.

Fig.11 Miller Audio Research Jitter Analyzer, waveform of Fs/4 test data (4 data points/cycle).

Fig.12 Musical Fidelity X-24K, output waveform when decoding fig.11 data.

However, fig.13 shows what comes out of the MSB's output when the 4x oversampling filter is switched off. Instead of the smooth sinewave seen in fig.12, there is a staircase wave, with 32 distinct sample points evident per cycle. (Remember, there were only four data points per cycle in the original data.) It looks as if the data are being oversampled by a factor of eight, but no reconstruction filter is being used. (The sharply defined steps in the waveform imply the existence of content well above the signal's nominal passband of 22.05kHz.)

Fig.13 MSB Platinum Link Plus, 4x Upsampling turned off, output waveform when decoding fig.11 data.

This waveform was not affected by pushing the Platinum's front-panel buttons marked "16x Interpolation" or "Brickwall Filter On." However, switching in "4x Upsample" gave the waveform shown in fig.14. The number of sample points per waveform cycle has indeed increased by a factor of four, to 128, giving a better representation of the sinewave. However, it still doesn't approach the purity of the sinewave produced by the conventional combination of digital oversampling filter and analog low-pass filter seen in fig.12, and again, the sharp corners of the waveform imply the existence of ultrasonic content above the nominal passband.

Fig.14 MSB Platinum Link Plus, 4x Upsampling turned on, output waveform when decoding fig.11 data.

I have no idea whether this behavior is just a quirk or is essential to getting the best sound quality from the MSB Platinum Link Plus. But I was certainly surprised by it!—John Atkinson

14251 Pescadero Road
La Honda, CA 94020
(650) 747-0400