Media Server Reviews

When I reviewed the $4250, Ethernet-connected Mirage MMS-5A media server in the April 2013 issue, I had been given to understand that while its four pairs of analog outputs offered only utilitarian performance (which I confirmed in my review), its 24-bit/192kHz–capable S/PDIF output was "audiophile-grade." I was disappointed, therefore, to discover that the Mirage MMS-5A's S/PDIF output was not bit-accurate. With 16-bit data and the MMS-5A set to output 16-bit data with no sample-rate conversion, it appeared to add noise at the LSB level—dither, perhaps?—that reduced resolution. Autonomic recommends that the MMS-5A's digital output be set to the highest possible bit depth and sample rate for the best sound quality. It was only when I reset the Mirage's output to 24/192, using the Config-Source page on the Web browser interface, that I got the expected 16-bit resolution. But because the Mirage was now upsampling the data, it was no longer bit-accurate, especially with 24-bit data, where resolution appeared to be lost.

After the April issue had gone to press, I heard from Autonomic that they weren't defining the S/PDIF output as "audiophile-grade," which is what I had understood from the MMS-5A's promotional literature and manual. Instead, the "audiophile-grade" outputs were the USB ports on the Mirage's embedded Intel motherboard: "The asynchronous USB output delivers a pure data stream of jitter-free 192/32-bit digital audio designed to be used with any outboard DAC."

There are two pairs of USB jacks on the Mirage MMS-5A's rear panel. The yellow ones are USB 2.0 and the blue are USB 3.0, but all can be used either to connect external hard drives or to output audio data. Obviously, a Follow-Up review was necessary, to assess the MMS-5A via its USB ports.

I was puzzled by Autonomic's referring to the USB port as "asynchronous." My understanding is that when a USB connection is used for an asynchronous flow of data, control of the sending of those data is handed over to the receiving device, usually a D/A processor of some kind. This ensures that timing uncertainty in the datastream is kept to a minimum. However, the "asynchronous" nature of the USB connection is a property of the receiving device or "sink," not the source PC.

For this Follow-Up, I intended to use four outboard D/A processors with the Mirage: the AudioQuest DragonFly (reviewed in October 2012); the Halide DAC HD (August 2012); the latest version of the Musical Fidelity M1DAC (March 2011); and the Musical Fidelity M6DAC (to be reviewed in the June issue). All of these DACs have true asynchronous USB ports. However, while the Musical Fidelity M1, Halide, and AudioQuest worked correctly when connected to one of the MMS-5A's USB ports—each was identified as "unknown USB DAC" in the Web browser's Source selection page—the Musical Fidelity M6 couldn't be recognized, and I got an error message. (While the M1DAC, Halide, and AudioQuest use the TI TAS1020B USB 1.1 data receiver chip, the M6DAC uses an XMOS USB 2.0 chip, if that's relevant.) In addition, though the DragonFly, DAC HD, and M1DAC will all work at a sample rate of 88.2kHz, the only sample-rate options available with the Config-Source page were 44.1, 48, and 96kHz, all with a word length of 24 bits. The M6DAC will operate at up to 192kHz via USB, but that was rendered moot by its apparent incompatibility with the MMS-5A.

Before auditioning the Mirage, I examined its behavior feeding USB data to the AudioQuest and Musical Fidelity M1 DACs 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"). I have no way of directly examining datastream jitter with a USB connection, so I looked at the spectra of the DACs' analog outputs while they handled the 16/44.1 Miller-Dunn J-Test data (a high-level tone at 1/4 the sample rate on which has been superimposed an LSB-level squarewave at 1/192 the sample rate).

To set a baseline for the effect of datastream jitter, fig.1 shows the spectrum of the M1DAC's analog output decoding 16-bit J-Test data sourced via USB from a MacBook Pro. The regular series of spikes in the spectrum are the residual odd-order harmonics of the low-frequency squarewave; these are at the correct level, and are not being accentuated either by the USB connection/receiver chip or by the D/A circuitry. There is a pair of sidebands at ±120Hz, which means that these are in some way related to the power supply.

Fig.2 shows the spectrum of the M1DAC's analog output fed the same 16-bit J-Test data from one of the Mirage MMS-5A's USB 2.0 ports. It looks very different, but on close inspection, the levels of the odd-order harmonics of the low-frequency squarewave are actually correct. The sidebands at ±120Hz are absent, but now a complete series of even-order harmonics is visible. The MMS-5A had been set to 24/44.1 for this measurement; I reset it to 24/96, but there was no change in the measured result, nor was there when I tried sourcing the data from one of the Mirage's USB 3.0 ports. I got identical results with the Halide DAC HD, so this anomalous behavior will be due to the Mirage, not to the DAC with which it is being used.

It is impossible to find out the reason for this unusual behavior. Perhaps something is being done to the data at the LSB level that results in even-order distortion, which typically results from some kind of signal asymmetry. However, looking at the waveform of an undithered 16-bit signal at exactly –90.31dBFS, as reproduced by the Musical Fidelity M1DAC when fed data from the Mirage MMS-5A via USB (fig.3), suggests that nothing is being done to the data. The three DC voltage levels are clearly resolved, and the Gibbs Phenomenon "ringing" introduced by the Musical Fidelity's reconstruction filter is also visible.

Finally, as I mentioned in my April review, the 24-bit data output from the MMS-5A's S/PDIF port appeared to be truncated. To examine this aspect of the Mirage's behavior via its USB ports, I played WAV files representing a dithered 1kHz tone at –90dBFS, with word lengths of first 16 and then 24 bits, feeding the data via USB to the AudioQuest DragonFly. The cyan and magenta traces in the spectra in fig.4 were taken with 16-bit data; the tone peaks correctly at –90dBFS, no distortion harmonics are visible, and the noise floor lies just above the 16-bit level. However, while the noise floor drops to the DragonFly's residual level with 24-bit data (blue and red traces), distortion harmonics are now visible at 3, 5, 7, and 9kHz. That these harmonics are not seen in the DragonFly's analog output when fed 24-bit USB data from other sources (see fig.4 here) suggests that bit-depth truncation is taking place somewhere in the datapath inside the Mirage.

These measurements, as well as the fact that data representing a 20kHz tone sampled at 44.1kHz output from the Mirage MMS-5A's USB port has the same decrease in level that I found from its S/PDIF port, regardless of the DAC being used, suggests that Autonomic's USB port is still not transparent to the audio data passed through it. Yes, it does handle 16-bit data correctly, unlike the S/PDIF port, in that it does not add noise in the digital domain—but it still does not appear to pass 24-bit data correctly. Until this is cleared up, most likely with a change in firmware (the review sample was running version 4.0.12150.1), I must withhold recommendation for what is otherwise an attractive product with a well-sorted-out user interface.—John Atkinson