Hi-Rez Disc Player/Transport Reviews

Sidebar 3: Measurements

I measured the Playback Designs MPS-5 using the Audio Precision SYS2722 system (see www.ap.com and "As We See It" in the January 2008 issue), as well as, for some tests, my Audio Precision System One Dual Domain and the Miller Audio Research Jitter Analyzer. To examine the performance of the player's USB input, I drove it with the USB 2.0 output of a Shuttle PC running Windows Vista, playing WAV files using Foobar 2000 with the WASAPI plug-in to give bit-accurate playback. The MPS-5's AES/EBU data input successfully locked to datastreams with sample rates ranging from 32 to 192kHz; the TosLink input, however, appeared to be limited to a maximum of 96kHz. The USB input, which uses a Burr-Brown PCM2902 receiver chip, was limited to rates up to 48kHz and 16-bit data.

Until the MPS-5, the best error correction I had encountered was from the dCS Puccini (reviewed in December 2009, which also uses an Esoteric mechanism), which didn't suffer from glitches in its audio output until the gaps in the data spiral of the Pierre Verany Test CD reached 3mm in length. The MPS-5 exceeded that, playing through 3.5mm gaps in the data, and not suffering from occasional mutes until the gaps reached 4mm. Astounding performance!

The player's maximum output level was 4.12V from the balanced XLR jacks, and half that figure, as expected, from the unbalanced RCA and BNC outputs. The output impedance was very low and uniform with frequency from all outputs, at 22 ohms (balanced), 11 ohms (unbalanced RCA), and 45 ohms (unbalanced BNC). The last is 10% lower than the specified 50 ohms required to optimally drive the darTZeel NHB-18NS preamplifier's BNC input. With all data sources, all of the MPS-5's outputs preserved absolute polarity; ie, were non-inverting.

With external PCM data at 96 and 192kHz, the MPS-5's frequency response (fig.1) started rolling off in the top audio octave, reaching –0.5dB at 20kHz and –3dB at 32kHz (96kHz data), and 51kHz (192kHz data), which is more band-limited than other SACD players. Peculiarly, while the channel balance was perfect with high-sample-rate data, with CD playback, the right channel (fig.1, gray trace) was 0.2dB below the left (green). Playing back the "provisional" Sony test SACD, the player's output was down by the same 0.5dB at 20kHz, but the ultrasonic response rolled off at the same rate it had done with 96kHz PCM, reaching –3dB at 32kHz and –13.4dB at 50kHz (fig.2). Playing back a preemphasized CD, the MPS-5's response was the same as with normal data (not shown), suggesting negligible deemphasis error. Channel separation was excellent, at >110dB below 1kHz, decreasing to a still-good 85dB at 50kHz (not shown).

Fig.1 Playback Designs MPS-5, frequency response at –12dBFS into 100k ohms at: 192kHz sample rate (left channel cyan, right magenta), 96kHz (left blue, right red), 44.1kHz (left green, right gray). (1dB/vertical div.)

Fig.2 Playback Designs MPS-5, SACD frequency response at –3dBFS into 100k ohms (left channel blue, right red). (1dB/vertical div.)

As usual, for reasons of consistency with Stereophile's library of digital-product measurements, which now goes back more than 20 years, my first test for resolution is to analyze the spectrum of the output signal of the device under test while it plays data representing a dithered 1kHz tone at –90dBFS, using a swept 1/3-octave bandpass filter. The result with 16-bit CD data is shown as the top pair of traces in fig.3. Usually, other than the peak at 1kHz touching the –90dB line, these traces show only the recorded dither noise. The Playback player, however, did appear to be contributing some noise, and to a greater extent in the left channel (solid trace) than in the right (dashed). There was also a trace of 120Hz hum apparent in the left channel, though at –113dB, this will not be audible.

Increasing the bit depth to 24 or playing back DSD data usually drops the noise floor low enough to reveal the player's or processor's own noise floor. (For an example of superb performance with this test, see the measurements accompanying the review of the Bryston BDA-1 processor elsewhere in this issue.) But to my surprise, given the Playback Designs player's pedigree, both 24-bit PCM and DSD data gave a noise floor that was only 3dB lower than with CD data (fig.3, bottom two pairs of traces). And again, the left channel was noisier than the right—so much so that the left channel with DSD and 24-bit data was noisier than the right channel with CD data. Repeating the spectral analysis using an FFT technique confirmed the MPS-5's disappointing performance (fig.4), with the noise almost high enough in level to obscure a dithered tone at –120dBFS. In addition, the noise floor for SACD playback was disturbed by some low-level enharmonic spikes (fig.5).

Fig.3 Playback Designs MPS-5, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS with: 16-bit data (top), 24-bit data (middle at 2kHz), DSD data (top trace at 20kHz). (Right channel dashed.)

Fig.4 Playback Designs MPS-5, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS with: 16-bit data (left channel cyan, right magenta), 24-bit data (left blue, right red).

Fig.5 Playback Designs MPS-5, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS with DSD data on SACD (left channel blue, right channel red).

Was the analysis being corrupted by ultrasonic noise being folded back into the audioband? Apparently not, as playing a CD track with a –1LSB DC signal gave the 1/3-octave spectral analysis shown in fig.6. Again, the left channel is noisier than the right, and while the ultrasonic noise from the DSD encoder's noiseshaping gives a noise floor that rises above the audioband, this is not extreme (fig.6; note that the rise in noise is exaggerated by higher-frequency components leaking past the bandpass filter's skirts). Looking at the output signal on an oscilloscope seemed to reveal both steady-state high-frequency noise and random bursts of higher-frequency noise. So while checking linearity error with spot tones gave excellent results down to –105dBFS or so, the result of my usual continuous sweep, which takes about 20 seconds to complete, was spoiled by bursts of noise (not shown).

Fig.6 Playback Designs MPS-5, 1/3-octave spectrum with noise and spuriae of digital black, 16-bit data. (Right channel dashed.).

The high-frequency noise obscures the shape of an undithered 16-bit PCM sinewave at –90.31dBFS (fig.7). You can hardly make out the three DC levels that you can see in the measurements sidebar for the Bryston BDA-1. And, as suggested by figs. 3 and 4, the picture is only a little better with DSD data (fig.8).

Fig.7 Playback Designs MPS-5, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit data (left channel blue, right red).

Fig.8 Playback Designs MPS-5, waveform of dithered 1kHz sinewave at –90dBFS, DSD data (left channel blue, right red).

The MPS-5 had low levels of harmonic distortion, even into the demanding 600 ohm load (fig.9), and only the subjectively benign second and third harmonics rise significantly above the level of background noise. (This graph was plotted using 24-bit PCM data; again the level of background noise is closer to 16-bit performance. DSD data didn't give a result that was appreciably better.) Intermodulation distortion was also very low, with all the distortion and aliasing products resulting from an equal mix of 19 and 20kHz tones at or below –94dB, or 0.002% (fig.10).

Fig.9 Playback Designs MPS-5, spectrum of 50Hz sinewave at 0dBFS into 600 ohms, 24-bit data (left channel blue, right red; linear frequency scale).

Fig.10 Playback Designs MPS-5, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 600 ohms, 24-bit data (left channel blue, right red; linear frequency scale).

Finally, playing the Miller-Dunn diagnostic jitter test tone from a CD, the MPS-5 gave a very low level of word-clock jitter, the Miller Analyzer indicating just 123 picoseconds peak–peak, which is actually at the analyzer's resolution limit. When I fed the MPS-5 external 16-bit data from my PC via 15' of TosLink—very much a worst-case situation—the jitter rose to a still quite low 457ps, primarily due to a pair of sidebands at the data-related frequencies of ±229Hz. With the MPS-5 fed the same data via USB, the measured jitter level was unmeasurable with the Miller Analyzer. Though data-related sidebands can be seen in the right channel (fig.11), these are at the residual level of the squarewave harmonics in the test signal.

Fig.11 Playback Designs MPS-5, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit USB data. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz (left channel blue, right red).

Although I did not include the measurment in the original publication of the review, I did check the MPS-5's impulse response (fig.12). It looks as if the MPS-5 is using two filters in series: one with good time-domain behavior like the Wadia "Ultramaster," which had exactly the same half-cycle of undershoot before and after the impulse but no ringing; and then a minimum-phase low-pass to reduce the level of the residual ultrasonic images. Using two low-pass filters in series would explain the premature ultrasonic rolloff I found.

Fig.12 Playback Designs MPS-5, impulse response (a single sample at 0dBFS).

Early on in the testing of the Playback Designs MPS-5, worried that there was something wrong with our review sample, I took the cover off to check that all the ribbon cables were seated properly (they were) and that there was nothing obviously adrift (there wasn't). So while I was impressed by the player's standard of construction, I can't say the same about its technical performance. The relatively high level of background noise limits the MPS-5's resolution with SACD and external 24-bit data to not much better than 16-bit CD. I am puzzled, therefore, why Michael Fremer liked the sound of this player so much. Perhaps his description of its sound being "analog-like" is a clue—for reasons that are not fully understood, a signal with very-low-level random noise added is sometimes preferred, on that it is more intelligible, to the same signal without such noise.1 But I feel that the MPS-5's measured performance precludes an unreserved recommendation.—John Atkinson

---

Footnote 1: See, for example, "Stochastic Resonance in Acoustic Emission," M. Friesel, Journal of Testing and Evaluation, 1999, and "The Benefits of Background Noise," Moss, Wiesenfeld, Scientific American, August 1995.