Benchmark DAC1 USB D/A processor & headphone amplifier Measurements
I examined the Benchmark's measured behavior using three different test systems: the National Instruments–based Miller QC Suite, my vintage Audio Precision System One Dual Domain, and a loaned sample of Audio Precision's top-model SYS2722 system (see this issue's "As We See It.").
Looking first at the DAC1 USB's performance when fed data via its conventional AES/EBU and S/PDIF inputs, it locked to data with sample rates ranging from 32 to 192kHz (though the frequency response with 192kHz data didn't appear to be any higher than with 96kHz data). The maximum output level from the balanced outputs, with the internal jumpers set to "0dB," was a high 17.03V. The three 10dB pads each gave a 10dB reduction from the calibrated output level, as specified. These pads didn't affect the unbalanced output, which had a maximum output level, as supplied, of 2.014V. Also as supplied, the headphone output had a maximum output of 2.78V. (This can be increased by 10dB by removing internal jumpers adjacent to the jacks.) The output impedance from the headphone jacks was extremely low, at <0.5 ohm including 6' of interconnect. It was higher from the RCA jacks, but still usefully low at 30 ohms, while without any padding, the balanced jacks offered a 60 ohm impedance across the band. This rose to a maximum of 407 ohms with the –10dB pad in place, which is still fairly low. All the outputs preserved absolute polarity (ie, were non-inverting), the XLRs being wired with pin 2 "hot."
With the outputs set to "fixed," the channel matching was superb at 0.02dB or better. A slight, 0.4dB imbalance appeared when the volume control was in operation. This can be seen in fig.1, which plots the response from the balanced jacks with the control at its maximum with the DAC1 fed 44.1kHz data (magenta and green traces) and 96kHz data (blue and red traces). With both sample rates, the Benchmark's audioband response is flat, with an insignificant –0.2dB droop evident at 20kHz. The 96kHz data response continues the rolloff, to reach –1dB at 43kHz, before the usual precipitous rolloff by half the sample rate. Channel separation (not shown) was superb at >110dB at low frequencies.
Fig.1 Benchmark DAC1 USB, frequency response at –12dBFS into 100k ohms, 96kHz data (blue, left; red, right) and 44.1kHz data (magenta, left; green, right). (0.5dB/vertical div.)
To keep a basis of comparison with my library of measured performance, I first looked at the DAC1's resolution by sweeping a 1/3-octave bandpass filter down from 20kHz to 20Hz while it decoded dithered 16-bit data representing a 1kHz tone at –90dBFS. The result, shown as the top pair of traces in fig.2, really shows only the recorded dither noise. Increasing the bit depth to 24 gave the middle traces in fig.2; the peak still correctly touches the –90dB line, but now the noise floor is 20dB lower across the band, suggesting that the Benchmark's ultimate dynamic-range capability is better than 19 bits, which is superb. A small peak can now be seen centered on the AC supply frequency of 60Hz, but at –140dB, this is of academic interest only. Changing the data to represent a dithered 24-bit tone at –120dBFS gave the bottom pair of traces in this graph. Some negative linearity error can be seen in the left channel, as well as a peculiar if low-level bump around 750Hz.
Fig.2 Benchmark DAC1 USB, 1/3-octave spectrum with noise and spuriae of (from top to bottom at 3kHz): dithered 1kHz tone at –90dBFS with 16-bit and 24-bit data; dithered 1kHz tone at –120dBFS with 24-bit data (right channel dashed).
High-resolution FFT analysis of the DAC1's output while it decoded a 24-bit/1kHz tone at –90dBFS (fig.3) reveals that all the harmonic spuriae are at or below –136dB, but there is a hint of the same bump at 750Hz, as well as another just below 2kHz. Perhaps these are idle tones of some kind; their absolute level is so low at –140dB that they will be inconsequential.
Fig.3 Benchmark DAC1 USB, FFT-derived spectrum of dithered 1kHz tone at –90dBFS with 24-bit data (blue, left; red, right; linear frequency scale).
The red trace in fig.4 plots the left channel's linearity error using 24-bit data. Any error is negligible down to below –100dBFS, but the negative error below that level appears to correlate with that seen with the –120dB tone in fig.2. Nevertheless, it will be inconsequential, and the Benchmark's reproduction of an undithered tone at exactly –90.31dBFS (fig.5) was perfect, with excellent waveform symmetry, and clear delineation of the three DC voltage levels described by this signal. With 24-bit data, the waveform is a pretty good sinewave (fig.6).
Fig.4 Benchmark DAC1 USB, left-channel departure from linearity, 24-bit data (red, 2dB/vertical div.).
Fig.5 Benchmark DAC1 USB, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit data.
Fig.6 Benchmark DAC1 USB, waveform of undithered 1kHz sinewave at –90.31dBFS, 24-bit data.
Even at its maximum output level, the DAC1 had very low levels of harmonic distortion. With the balanced outputs driving the Audio Precision's 200k ohm load with a 24-bit tone at 0dBFS (fig.7), the subjectively benign second harmonic is the highest in level, though this lies at just –102dB left and –107dB right. The third and fourth harmonics are the only other components that rise above the –120dB line (0.0001%). This superb linearity could not quite be maintained at the balanced output's maximum level setting into 600 ohms, but I think it safe to say that being asked to deliver 17V into 600 ohms is not likely to happen in the real world. Intermodulation distortion was also very low, with all the products resulting from an equal mix of high-level 19 and 20kHz tones lying at –104dB or lower (fig.8).
Fig.7 Benchmark DAC1 USB, spectrum of 1kHz sinewave at 0dBFS into 100k ohms (blue, left; red, right; linear frequency scale).
Fig.8 Benchmark DAC1 USB, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms (blue, left; red, right; linear frequency scale).
The circuit topology used by Benchmark should show a high level of immunity to word-clock jitter, and spectral analysis of its output while it decoded a 16-bit version of the Miller J-Test signal (fig.9) is dominated by the residual odd-order harmonics of the low-frequency squarewave rather than jitter. A pair of sidebands can be seen at ±1500Hz, but these are low in level, and the Miller Analyzer reported just 157 picoseconds peak–peak of jitter. Increasing the word length to 24 bits (fig.10) dropped the squarewave harmonics below the graph's floor, leaving just a few pairs of sidebands. The measured jitter level was now just 119ps p–p, which is both superbly low and at the limit of the Miller Analyzer's resolution. Note also that both of these measurements were taken using a 15' length of TosLink optical cable from the S/PDIF output of the RME soundcard fitted to an older PC in my test lab, which is very much a worst case.
Fig.9 Benchmark DAC1 USB, high-resolution jitter spectrum of analog output signal (11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz), 16-bit data sourced from PC via 15' TosLink. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.10 Benchmark DAC1 USB, high-resolution jitter spectrum of analog output signal (11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz), 24-bit data sourced from PC via 15' TosLink. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Turning to the DAC1 USB's behavior when driven by its USB input, I initially used Bias Peak 5.2 running on my 2002-vintage Apple PowerBook (OSX 10.3.9, USB 1.1) to play back 24-bit files from my test-tone library. (Bias Peak has its own OSX CoreAudio dialog to ensure that the operating system passes the data intact out through the selected device without applying sample-rate conversion or any unwanted equalization or level changes. I had previously validated the program's bit transparency with both Echo Indigo PC Card and Metric Halo FireWire devices.) The Benchmark did indeed decode 24-bit/96kHz sampled data through its USB connection, and in all respects, its measured performance was identical to that through the AES/EBU and S/PDIF inputs. Fig.11, for example, shows the FFT-derived spectrum of its balanced analog output while it decoded a dithered 24-bit tone at –90dBFS; it is identical to that shown in fig.3, which was taken using TosLink data.
Fig.11 Benchmark DAC1 USB, FFT-derived spectrum of dithered 1kHz tone at –90dBFS with 24-bit data, sourced from Apple TiBook via USB connection (blue, left; red, right; linear frequency scale).
For comparison with fig.2, the top trace in fig.12 shows a 1/3-octave analysis of the DAC1 USB's output with the same 24-bit data fed via USB. It is indeed the same as the middle pair of traces in fig.2. But the bottom pair of traces shows what happened when I tried to repeat the test with dithered 16-bit data representing the same signal fed to the DAC1 via USB. The processor appears to have gone deaf; all that can be seen in the output of the 1/3-octave bandpass filter is the analog noise floor!
Fig.12 Benchmark DAC1 USB, 1/3-octave spectrum with noise and spuriae of (from top to bottom at 3kHz): dithered 1kHz tone at –90dBFS with 24-bit and 16-bit data sourced from Apple TiBook via USB connection (right channel dashed).
This was both unusual and unexpected, to say the least. If the Benchmark can correctly decode 24-bit data via a USB connection, it should also do so with 16-bit data. I repeated the analysis with tones at various levels, and it appeared that the output rapidly dropped to zero with a 16-bit sinewave at levels below –67dBFS.
There are three variables involved here besides the Benchmark: the software, the computer/operating system, and the flavor of USB. I therefore repeated the test using iTunes on the PowerBook, making sure that the Quicktime engine wasn't performing any sample-rate conversion and that all "enhancements" were disabled. The results were the same. I then connected the Benchmark to my lab PC (a dual-core Pentium running XP, fitted with USB 2.0 ports), followed Benchmark's thorough instructions about ensuring that Windows wasn't doing anything untoward, and played the same 16- and 24-bit files using Adobe Audition. Again, while the 24-bit data were handled correctly via the USB connection, the 16-bit files dropped out below –67dBFS. Windows Media Player won't handle 24-bit files, but trying that program with the 16-bit files again gave the same anomalous results. As did Winamp. My Mac mini running OSX 10.4.10 behaved correctly, however.
Was this an artifact of the measurement setup? I thought it unlikely, considering that I got the same results with three different test sets. However, as one final investigation, I created a 16-bit test signal that combined a high-level tone, 19kHz at –10dBFS, with a low-level one, 1kHz at –70dBFS. The latter level was below the point at which the 1kHz tone on its own had disappeared using the PowerBook as source, but when mixed with the high-level tone, it was still present in the Benchmark's output. This is an enigma, considering I had no problem sending 24-bit data to the Benchmark via the PowerBook, or 24- and 16-bit data to the DAC1 from my Mac mini.
I stopped the testing at that point, as deadlines loomed. I will return to this matter in a Follow-Up review. In the meantime, it is possible either that my sample of the DAC1 USB has a problem handling 16-bit USB data (unlikely), or that I have a problem setting up PC or a Mac running OSX 10.3.9 to send 16-bit data via USB to the DAC1 (quite likely).—John Atkinson