Benchmark DAC2 HGC D/A processor/headphone amplifier Measurements
Sidebar 3: Measurements
I examined the Benchmark DAC2 HGC'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"). The DAC2 HGC's performance via its USB port was tested using my 2012-vintage Apple MacBook Pro. Apple's USB Prober utility identified the Benchmark as having the Product String "Benchmark DAC2 USB Audio 1.0" or "Benchmark DAC2 USB Audio 2.0," depending on which mode the DAC2's USB input had been set to. The Benchmark operated with the preferred isochronous asynchronous protocol in both USB modes. Set to USB1.0, the DAC2 accepted data with sample rates up to 96kHz; set to USB2.0, it also worked with data sampled at 176.4 and 192kHz. As usual, the TosLink input locked to data at sample rates of up to 96kHz, while the coaxial input also successfully locked to 176.4 and 192kHz datastreams. All the measurements were taken from the rear-panel analog outputs rather than the headphone outputs.
When the DAC2 was first turned on, the volume control smoothly rotated itself so that the level was 7dB lower than the maximum. The maximum output level at 1kHz was 3.47V from the balanced XLR jacks, and exactly half that, as expected, from the single-ended RCA jacks. (The 10dB pads were installed on the XLR outputs.) This was with S/PDIF data; to my surprise, the USB input produced an output 2.2dB higher in level. Both outputs preserved absolute polarity (ie, were non-inverting), with the front-panel Polarity LED dark, meaning that the XLRs are wired with pin 2 hot. The output impedance was 407 ohms from the XLRs and 30 ohms from the RCAs, from 20Hz to 20kHz.
The Benchmark's impulse response at 44.1kHz (fig.1) indicates that its reconstruction filter is a conventional, time-symmetrical, linear-phase type. The cyan and red traces in fig.2 are a wideband spectral analysis of the DAC2's output while it was fed 44.1kHz data representing white noise at 4dBFS. The signal rolls off steeply above the audioband, reaching the noise floor just below 23kHz. As a result, the sampling-generated image at 25kHz of a full-scale 19.1kHz tone (blue and magenta traces) is completely suppressed, while the second and third harmonics of the tone, at 38.2 and 57.3kHz, lie at or below 100dB (0.001%). (This graphical representation of the reconstruction filter's behavior was suggested to me by Jürgen Reis of MBL.) Fig.3 shows the Benchmark's frequency response with data sampled at 44.1, 96, and 192kHz. In each case, the response is flat until just below half the sample rate. Channel separation via the digital inputs (not shown) was superb, at >125dB in both directions below 1kHz, and still 100dB at 20kHz.
The Benchmark DAC2 offered one of the highest resolutions I have measured. The cyan and magenta traces in fig.4 show the spectrum of the processor's XLR outputs with its volume control set to its maximum while it decoded dithered data representing a 1kHz tone at 90dBFS. Other than the spike at 1kHz, which peaks at the correct level, the traces show only the spectrum of the dither noise used to encode the system. Increasing the bit depth to 24 gave the blue and red traces in this graph. Note that I've had to extend the vertical scale to 160dBFS to reveal the noise floor. The drop in the level of the noise floor with 24-bit data indicates that the DAC2 offers 21-bit resolution, which is state-of-the-art performance. This graph was taken with TosLink data; the DAC2 behaved just as superbly via its USB port, but still offered some very low-level harmonics, predominantly in the left channel (blue trace).
With the very low level of noise and superb linearity, the DAC2's reproduction of an undithered 16-bit tone at exactly 90.31dBFS was essentially perfect (fig.5): the waveform is symmetrical about the time axis, the three DC voltages described by the data are well resolved, and the reconstruction filter's symmetrical Gibbs Phenomenon "ringing" on the tops and bottoms of the waveform is readily evident. With undithered 24-bit data, the result is an excellent sinewave (fig.6).
Even driving a very demanding 600 ohm load, the DAC2 offered very low levels of analog distortion (fig.7), with the third harmonic the highest in level, at 116dB (0.00015%). Other than the second and fifth, all the other harmonics lie at close to 130dB. Intermodulation distortion was also superbly low, again even into 600 ohms (fig.8).
Tested for its rejection of word-clock jitter via its TosLink input, which is the worst case, a 16-bit version of the Miller-Dunn J-Test data resulted in a spectrum almost completely free from jitter-related sidebands (fig.9), and with the odd-order harmonics of the low-frequency, LSB-level squarewave all at the correct levels. With 24-bit J-Test data (fig.10), other than the very low-level sidebands at ±1500Hz in the left channel (blue trace), no jitter-related spuriae at all are visible! To my surprise, however, when I repeated the 24-bit test via the DAC2's USB port, a pair of sidebands appeared at ±229Hz (fig.11), though 16-bit J-Test data via USB gave a spectrum identical to that in fig.10. This graph was taken with the port set to USB2.0; repeating the test with the port set to USB1.0 gave an identical result. The J-Test signal is not diagnostic for USB data, as the bit clock is not embedded in the data, as it is in S/PDIF and AES/EBU datastreams. The data-related sidebands in fig.11 are unlikely to be jitter-related, therefore. But something is happening with 24-bit USB dataperhaps this is the Logic-Induced Modulation reported by Meitner and Gendron in their October 1991 AES paper.
Used as an analoganalog preamplifier, the DAC2 HGC offered an unbalanced input impedance of just under 19k ohms at 20Hz and 1kHz, this dropping slightly to 17.4k ohms at 20kHz. (There are no balanced analog inputs.) The maximum gain was 6.4dB from the balanced outputs, 0.6dB from the unbalanced; both outputs preserved absolute polarity. The analog inputs offered a very wide frequency response that was not affected by reducing the load impedance from 100k ohms (fig.12, blue and red traces) to 600 ohms (cyan, magenta). The channels are superbly matched in this graph, both dropping by only 0.6dB at 200kHz. Reducing the volume didn't affect the DAC2 HGC's ultrasonic bandwidth, but a slight imbalance of 0.25dB in favor of the left channel appeared at the control's 1 o'clock position.
I measure a preamp's signal/noise ratio in the worst-case condition: with the input shorted but the volume control set to its maximum, Even so, the wideband, unweighted S/N, ref. 1V, was a superb 95.8dB in the left channel and 90.7dB in the right. Restricting the measurement bandwidth to the audioband improved both channels' ratios to 113.6dB; A-weighting gave a further improvement, to 116.4dB. Regarding channel separation and harmonic and intermodulation distortion for the analog inputs, the results were just as superb as for the digital inputs, so I haven't shown them.
Summing up the Benchmark DAC2 HGC's measured performance is easy: It's simply superb.John Atkinson