Media Server Reviews

I examined the Marantz Reference NA-11S1's electrical performance with 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"); for some tests, I also used my vintage Audio Precision System One Dual Domain. Because of the NA-11S1's complexity, I performed most of the testing using its TosLink input, repeating some of the tests using: test files stored on a USB stick, streamed to the Marantz via USB from my MacBook Pro mini running Pure Music; files pulled from my Mac mini running Twonky Server over a hardwired network; and files played from my MacBook Pro using AirPlay. The NA-11S1's DC filter was disabled for the measurements.

Connected to my MacBook Pro via USB, the Marantz was identified as the "NA-11S1" from "D & M Holdings Inc.," and as a "USB High Speed Device" operating in "Isochronous asynchronous" mode. Sample rates handled by the rear-panel USB port range from 32 to 352.8kHz but, peculiarly, not 384kHz. The front-panel USB port was restricted to sample rates of 96kHz and below.

The maximum output level at 1kHz was 4.76V from the balanced XLR jacks, 2.36V from the unbalanced RCA jacks. The balanced outputs inverted signal polarity, meaning that the XLR jacks had their pin 3 wired as hot, not pin 2, as is usual. The unbalanced outputs preserved absolute polarity; ie, were non-inverting. The balanced output impedance was a low 93 ohms at high and middle frequencies, rising very slightly and inconsequentially to 110 ohms at 20Hz. The respective unbalanced impedances were 46.5 and 51.5 ohms.

Fig.1 shows the impulse response of the NA-11S1's default Filter 1. The impulse is inverted, as expected, and there is just a single cycle of ringing before and after the impulse, which correlates with a slow rolloff above the audioband (fig.2, red trace) and a null at 44.1kHz. In this kind of presentation, pioneered by Jürgen Reis of MBL, the blue trace shows the spectrum of the NA-11S1's output while it decodes 44.1kHz-sampled data representing a full-scale 19.1kHz tone. Not only is the level of this tone slightly suppressed by Filter 1, the aliasing product at 25kHz (44.1kHz–19.1kHz) is almost at the same level. Note, however, the very low levels of distortion harmonics associated with the 19.1kHz tone.

Filter 2's impulse response is shown in fig.3. While not minimum phase, there is actually less ringing before the impulse than after. Filter 2 has a much sharper rolloff than Filter 1 (fig.4, red trace), though the stop-band attenuation is not as great as is usually the case with linear-phase filters. However, the 19.1kHz tone is now reproduced at the correct level, and the aliasing product at 25kHz is suppressed by more than 50dB (blue trace).

The frequency responses with Filter 1 and sample rates of 44.1, 96, and 192kHz are shown in fig.5. The shape of the response is the same with all three rates, just transposed by the change in sample rate. With 44.1kHz data (gray and green traces), the rolloff starts above 6kHz and reaches –3dB at 20kHz. With the 96 and 192kHz rates, the response is flat within the audioband, allowing the listener to get the benefit of the optimized time-domain performance without any frequency-domain compromise. With Filter 2 (fig.6), the response is the same at all three sample rates, interrupted by a sharp rolloff beginning just below the Nyquist Frequency (half of each sample rate).

Note the slight channel imbalance (0.15dB with Filter 1, 0.25dB with Filter 2) in these two graphs, in favor of the left channel. Channel separation (not shown) was superb, at >115dB in both directions below 1kHz, decreasing, due to the usual capacitive coupling, to a still-good 89dB at 20kHz.

Tested via its TosLink input, the Marantz had impressive resolution. Increasing the bit depth from 16 to 24, with a dithered 1kHz tone at –90dBFS, drops the noise floor by 20dB (figs.7 & 8), implying resolution approaching 20 bits, which is close to the state of the art. Repeating the spectral analysis for the 24-bit tone via Ethernet and Twonky Server gave the same result (fig.9), as did repeating it for 24-bit data on a USB stick and 24-bit data streamed from my MacBook Pro via USB, indicating that the NA-11S1 doesn't compromise resolution via any of its physical inputs. However, repeating the spectral analysis with AirPlay data suggested that this mode truncates data with bit depths greater than 16 (not shown). As Marantz admits in the NA-11S1's manual, turning on the noise-shaping function had no effect on the measured resolution or audioband noise floor.

Linearity error with 16-bit data (not shown) was negligible to below –100dBFS, and with its low level of analog noise, the NA-11S1's reproduction of an undithered tone at exactly –90.31dBFS was essentially perfect (fig.10). The waveform is symmetrical, and the three DC voltage levels described by the data are well defined. Even at this very low signal level, the result with undithered 24-bit data is an excellent-looking sinewave (fig.11).

The Marantz NA-11S1 offered very low levels of harmonic distortion, with the second harmonic the highest in level, at just –106dB (0.0006%, fig.12). Intermodulation distortion was also extremely low, though Filter 1 (fig.13) offered much less suppression of the ultrasonic aliasing products than Filter 2 (fig.14). These three graphs were taken into a high 100k ohm load; dropping the load to a punishing 600 ohms didn't change the picture by very much.

The NA-11S1's rejection of word-clock jitter depended on the input used. Fig.15, for example, shows a narrowband spectral analysis of the Marantz's output while it reproduced 16-bit J-Test data fed to it via TosLink. The odd-order harmonics of the LSB-level, low-frequency squarewave are not accentuated, but sideband pairs of unknown origin can be seen at ±150 and ±745Hz. With 24-bit data, the odd-order harmonics disappear as expected, but the two sideband pairs remain (fig.16).

The picture was the same for coaxial S/PDIF data, but with the J-Test data played back from a USB stick, the sideband pairs now lie at ±260 and ±785Hz, with a third pair at ±1310Hz (fig.17). The best performance on the J-Test was with the data streamed from my laptop via USB (fig.18). The spike that represents the 11.025kHz tone is narrow, with no broadening visible at its base; the only sidebands visible lie at ±180Hz and are very low in level. With 24-bit Ethernet J-Test data (fig.19), the spectrum is very similar to that with the USB stick.

Overall, this is superb measured performance, though it looks as if the best performance—by a very small margin—is obtained when data are streamed from the computer via USB rather than via Ethernet or S/PDIF. I have no idea why that should be the case.—John Atkinson