Sidebar 3: Measurements
I measured the Aurender A10 with my Audio Precision SYS2722 system (see the January 2008 "As We See It"), using the Audio Precision's optical digital outputs, and data stored on the A10's internal hard disk and sourced from a NAS on my network. The serial number of the sample I'd been sent for measurement was ASA4A0096A10, its system software was v.5.10.16, and the Conductor app running on my iPad mini was v.2.8.6 (1730).
The optical inputs accepted data sampled up to 96kHz, the internal drive and the network connection files sampled at up to 384kHz. The A10's maximum output level at 1kHz was 4.04V from the balanced output jacks, and 2.02V from the unbalanced jacks with the output level set to Fixed or with the volume control set to its maximum. Both sets of outputs preserved absolute polarity, and the output impedance was an extremely low 1 ohm or lower at all audio frequencies, regardless of the impedance setting in the Conductor app (the options are Max current, Less current, and Min current). Channel separation was superb, at >130dB between 40Hz and 5kHz, and still 120dB or better at the frequency extremes. The A10's noise floor with 24-bit data was also very low, with no power-supply–related spuriae visible (fig.1). When I increased the bit depth from 16 to 24, the drop in the Aurender's noise floor with a dithered 1kHz tone at –90dBFS was around 21dB (fig.2), implying that the A10 offers resolution of close to 20 bits, which is superb. Consequently, the Aurender's reproduction of an undithered tone at exactly –90.31dBFS (fig.3) was close to perfect, with the three DC voltage levels described by the data clearly evident and the waveform superbly symmetrical.
The A10 offers five different reconstruction filters, as well as the MQA filter. I examined their behavior using S/PDIF data. Fig.9 shows the impulse response with data sampled at 44.1kHz for the Sharp Rolloff filter. It is a conventional finite impulse-response (FIR) type, with symmetrical ringing evident around the single sample at 44.1kHz. This filter's ultrasonic rolloff with 44.1kHz-sampled white-noise data is shown in fig.10 (red and magenta traces); the rolloff is very steep, and there is almost total suppression of the aliased tone at 25kHz associated with a full-scale tone at 19.1kHz (blue, cyan, footnote 1). Fig.11 shows the responses of this filter with data sampled at 44.1, 96, 192, and 384kHz. The ultrasonic rolloff conforms to the same basic shape up to 20kHz, with then a sharp rolloff evident at the lowest rate just below the Nyquist frequency (half the sample rate), but a more gentle rolloff at the three higher rates.
The e-mail exchange that followed was frustrating. Aurender's engineers in Korea didn't appear either to be able to repeat the problem or to comprehend the issue I was raising. I had no option but to postpone the review from the December 2017 issue to this one. Then, in mid-September, I received an e-mail from the US rep explaining that:
"We discovered that prior to applying the MQA firmware update earlier this year, MQA had recommended that we adopt using MQA up-sampling for all content in order to eliminate possible issues with click or pop noises when switching between non-MQA and MQA content. After some discussion with Alan at MQA about this, he had the following comment: 'The MQA decoder provides an optional up-sampler for PCM to simplify implementation and to enable a smooth, clean, click-free user experience. The reason this is offered is that the implementer may not know if the incoming stream is MQA and so the decoder is used to detect MQA and to provide a seamless switch to the usually higher output rate. By using Upsample Always, the user-experience is guaranteed to be accurate from the first sample of an MQA song and also to be free of clicks and pops if the user skips within a song or if there are cross-fades between songs.'
"As of our firmware update earlier this year, using the Upsample Always option in the MQA decoder is the current implementation in the A10. Alan's other comment was that when MQA upsampling is enabled you can still change the over sampling filter in the DAC during PCM playback, but as the DAC is being supplied 8x, the difference made by the different DAC filters will be harder to measure unless you can make a very fast digital capture.
"Our overall thinking is that for all intents and purposes, the filter selections are no longer valid, so we have removed the option to select the optional digital and analog filters from the Aurender Conductor App. We will be releasing a public update to the App to remove the filters on the Advanced tab so that the review can commence as planned."
Travel plans and other commitments meant that I couldn't get my sample of the A10 back on the test bench until the beginning of October 2017. When I did, and connected it to my network, the first thing the A10 did was to update its system software to the latest version, v.5.10.34. I resumed the testing and found that, despite what I had been told, the Filter options were still there in the Conductor app, which had also been updated to the latest version. The next thing I found was that that while the five different filters were still available for S/PDIF data, and measured identically to what I had found in my earlier testing, once again the MQA filter was incorrectly applied to non-MQA data when it was sourced from the A10's internal storage.
Its measured performance suggests that the A10's analog output stage is of high quality, and my experience with Aurender's N10 server makes me a fan of how the company's Conductor app organizes the user's music library. However, the A10's misapplication of the MQA reconstruction filter to non-MQA files stored on its internal drive means we must withhold a full recommendation for the A10 until this problem has been corrected.—John Atkinson
Footnote 1: My thanks to MBL's Jürgen Reis for suggesting this test to me. Footnote 2: See, for example, fig.9 here.
Fig.1 Aurender A10, spectrum with noise and spuriae of dithered 1kHz tone at 0dBFS with: 24-bit data (left channel blue, right red) (20dB/vertical div.)
Fig.2 Aurender A10, 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) (20dB/vertical div.).
Fig.3 Aurender A10, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit S/PDIF data (left channel blue, right red).
Harmonic distortion with a 50Hz tone at full level was very low. Even into the punishing 600 ohm load impedance (fig.4), the highest-level harmonic, the third, lay at –106dB (0.0003%). However, with a full-scale 1kHz tone, I was puzzled to see sidebands developing around the tone and its harmonics, even into the benign 100k ohm load (fig.5). These sidebands disappeared when I reduced the signal level by 3dB (fig.6) but not when I reduced the A10's output level by the same 3dB, so I wonder if they are due to mathematical limitations in the A10's digital signal processing rather than to power-supply limitations. Tested for intermodulation distortion with an equal mix of 19 and 20kHz tones, with the signal peaking at –3dBFS, the Aurender performed well, the difference product at 1kHz lying below –126dB and the higher-order products at –130dB (fig.7). However, even though this graph was taken with the factory-default reconstruction filter, labeled Short Delay, Sharp Rolloff, the aliased images of the two high-level tones are suppressed by just 12dB or so, as I would expect from a slow-rolloff filter (see later).
Fig.4 Aurender A10, spectrum of 50Hz sinewave, DC–1kHz, at 0dBFS into 600 ohms (left channel blue, right red; linear frequency scale).
Fig.5 Aurender A10, spectrum of 1kHz sinewave, DC–10kHz, at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.6 Aurender A10, spectrum of 1kHz sinewave, DC–10kHz, at –3dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.7 Aurender A10, Short Delay, Sharp Rolloff filter, HF intermodulation spectrum, DC–30kHz, 19+20kHz at –3dBFS into 100k ohms, 44.1kHz internal data (left channel blue, right red; linear frequency scale).
Tested for its rejection of word-clock jitter using 16-bit J-Test data sourced via TosLink, the A10 reproduced the odd-order harmonics of the low-frequency, LSB-level squarewave at the correct levels, as shown by the sloping green line in fig.8. However, a pair of sidebands of unknown origin can be seen at ±3.2kHz. The higher-frequency sideband also has some frequency smearing evident.
Fig.8 Aurender A10, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz: 16-bit TosLink data (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.9 Aurender A10, Sharp Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.10 Aurender A10, Sharp Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
Fig.11 Aurender A10, Sharp Rolloff filter, frequency response at –12dBFS into 100k ohms with S/PDIF data sampled at: 44.1kHz (left channel blue, right red), 96kHz (left gray, right green), 192kHz (left cyan, right magenta), 384kHz (left green, right gray) (1dB/vertical div.).
The Slow Rolloff filter has a very short impulse response (fig.12), with a high-frequency rolloff that starts around 12kHz, and very little suppression of the 25kHz aliased tone (fig.13). The Short Delay, SuperSlow Rolloff filter has an even shorter impulse response (fig.14). The spectrum with white noise has nulls at 44.1kHz and 88.2kHz and the output at 100kHz is down by just 30dB (fig.15). The Short Delay, Slow Rolloff filter is a minimum-phase type (fig.16) similar to Ayre Acoustics' Listen filter, with its ultrasonic rolloff similar to that shown in fig.11 (fig.17). The Short Delay, Sharp Rolloff filter has a conventional minimum-phase impulse response with 44.1kHz data (fig.18) with, as expected, a sharp ultrasonic rolloff (fig.19, red and magenta traces).
Fig.12 Aurender A10, Slow Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.13 Aurender A10, Slow Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
Fig.14 Aurender A10, Short Delay, SuperSlow Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.15 Aurender A10, Short Delay, SuperSlow Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
Fig.16 Aurender A10, Short Delay, Slow Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.17 Aurender A10, Short Delay, Slow Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
Fig.18 Aurender A10, Short Delay, Sharp Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.19 Aurender A10, Short Delay, Sharp Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
I then reexamined the behavior of the filters using the same test files stored on the A10's internal drive, and immediately ran into problems. The Short Delay, Sharp Rolloff filter still had a minimum-phase impulse response (fig.20), but it was much shorter than it had been with S/PDIF data. Its ultrasonic rolloff with 44.1kHz-sampled white noise (fig.21, red and magenta traces) was very different, and actually resembles the behavior of the MQA filters that I have measured in other processors (footnote 2). (This was why the A10's HF intermodulation result in fig.7, which I'd measured using internal data, was anomalous.) Fig.22 compares the wideband spectra of the A10's output when it decodes 44.1kHz-sampled white noise with S/PDIF data (green and gray traces) and internally stored data (blue, red). It looks as if the MQA filter is being applied to conventionally encoded PCM files when they are stored internally, but not when the A10 is decoding the same data via its S/PDIF input. Something was very wrong.
Fig.20 Aurender A10, internal data, Short Delay, Sharp Rolloff filter, impulse response (one sample at 0dBFS, 44.1kHz-sampled S/PDIF data, 4ms time window).
Fig.21 Aurender A10, internal data, Short Delay, Sharp Rolloff filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with 44.1kHz-sampled S/PDIF data (20dB/vertical div.).
Fig.22 Aurender A10, Short Delay, Sharp Rolloff filter, wideband spectrum of white noise at –4dBFS with internal data (left channel blue, right red) and S/PDIF data (left green, right gray), with data sampled at 44.1kHz (20dB/vertical div.).
I repeated all of this testing using data sourced from the NAS drive on my network. Again, regardless of which filter was being selected with the Conductor app, the MQA filter was being incorrectly applied to non-MQA data. I halted the testing and contacted Aurender's US representative, to let them know that there seemed to be something wrong with not only my sample of the A10 but also with Jason's.
Footnote 1: My thanks to MBL's Jürgen Reis for suggesting this test to me. Footnote 2: See, for example, fig.9 here.
