AudioQuest DragonFly USB D/A converter Measurements
I measured the AudioQuest DragonFly 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. I received two samples of the DragonFly: one from initial production, the second following a running production change in which a slight modification was made in the operation of the digitally controlled analog volume control.
With the DragonFly connected to one of the USB ports of my MacBook Pro, the Mac USB Prober utility reported the DragonFly as being the "AudioQuest DragonFly" manufactured by "AudioQuest inc.," and listed its serial number as "(C) 2011 Wavelength Audio, ltd." The last refers to Gordon Rankin's proprietary Streamlength asynchronous operating code for the TAS1020B USB receiver chip, and USB Prober confirmed that the DAC HD does indeed operate in the preferable isochronous asynchronous mode, with 24-bit word length and data sampled at 44.1, 48, 88.2, and 96kHz.
With the volume control set to its maximum with the MacBook's Control Panel, the maximum output level at 1kHz for both DragonFly samples was 1.86V, and their outputs preserved absolute polarity (ie, were non-inverting). The output impedance was very low, at 0.65 ohm including 6' of interconnect, and appropriate for driving headphones. Fig.1 shows the DragonFly's impulse response with 44.1kHz data. It indicates that the DragonFly uses a conventional linear-phase reconstruction filter. Fig.2 shows the frequency response with 44.1kHz (green and gray traces) and 96kHz (blue and red) data. The channels are superbly well matched, and the response is flat within the audioband with both sample rates. Channel separation (not shown) was good rather than great: 65dB in both directions across the audioband.
To remain consistent with the measurements of DAC resolution I've performed since 1989, I used a swept-bandpass technique to generate the traces in fig.3. The DAC is presented with dithered data representing a 1kHz tone at 90dBFS with either 16-bit words (top pair of traces) or 24-bit words (bottom traces). The traces peak at exactly 90dBFS, suggesting very low linearity error, which was confirmed with a separate test (not shown). Repeating the spectral analysis with a modern FFT technique gave the traces in fig.4. In both graphs, the increase in bit depth drops the noise floor in the upper midrange and treble by 10dB or so, suggesting that the DragonFly's ESS Sabre D/A chip has between 17 and 18 bits' worth of resolution. Considering that the DragonFly is powered from the 5V USB bus, this is good performance, and the noise floor is low enough not to obscure the shape of an undithered 16-bit waveform at exactly 90.31dBFS (fig.5). With 24-bit undithered data (fig.6), the DragonFly gives a good if rather noisy representation of a sinewave.
There was some modulation of the DragonFly's noise floor with signal level. The cyan and magenta traces in fig.7 show the level of the low-frequency noise floor with a full-scale 1kHz signal; with 1kHz tones at 60dBFS (green and gray traces) and 90dBFS (left, blue, right, red), the noise floor drops by up to 15dB. The increase in noise with the 0dBFS signal is presumably due to mathematical limitations in the digital circuitry, but is not as extreme as I've seen with some other DACs.
With the first sample of the DragonFly, a full-scale 24-bit signal actually clipped the bottom halves of the waveform with the computer's volume control set to its maximum, giving a THD+noise level of 3.8%. Backing off the control by one click (0.17dB) reduced the THD to 2.14%, by a second click (0.34dB) to 0.627%, and by a third click (0.51dB) to 0.054%, below which the THD+N percentage plateaued. The second sample didn't clip with a 0dBFS signal at maximum volume, and the THD+N was 0.041% rather than 3.8%. According to Gordon Rankin, the volume control offers 64 steps of less than 1dB to 60dB and then mute (100dB), but he used only 60 of those steps in the DragonFly, as the top four steps suffered from significant clipping into high impedances. "In retrospect," he wrote of the first sample, "I could have changed the maximum volume down a few more steps and then this would not have been an issue."
Basically, it looked as if the running change in production was to implement this suggestion of Rankin's, as the spectrum of the second DragonFly sample's output with a full-scale 1kHz signal at maximum volume was identical to that of the first sample's output set to 0.51dB (fig.8). The picket fence of low-level distortion harmonics in this graph indicates that the DAC is still on the verge of clippingbut with a 1kHz tone at 1dBFS, the higher harmonics have all dropped significantly in level, leaving the subjectively innocuous second harmonic the highest in level at a still-low 77dB (0.015%). (Ignore the presence of low-level AC supply-related tones in fig.9 compared with fig.8: for fig.9, I used my G4 Mac mini; for fig.8, my battery-powered laptop.) Even with the first sample, this lack of top-of-range linearity with the volume control at its maximum wouldn't be a problem with typical recordings, but something like the new Bonnie Raitt album, which peaks at 0dBFS pretty much all the time, might sound a little hard. The solution with the first sample would be to drop the level by three or more clicks, but with the second sample, this has been done for you. And, of course, if you're using the DragonFly as a headphone amplifier, where the volume control would not be used near the top of its range, this wouldn't be an issue with either sample.
The same issue raised its head in the high-frequency intermodulation test (fig.10). Even though the second sample was about to run out of dynamic range with the volume control set to its maximum, the 1kHz difference product still lay at an acceptably low 60dB (0.1%). This performance didn't appreciably worsen into low impedances, confirming the DragonFly's suitability for driving headphones.
The AudioQuest DragonFly offered excellent rejection of word-clock jitter. Even with a 24-bit version of the standard J-Test data, no signal-related sidebands were visible (fig.11). However, there was some slight widening of the skirts of the central spike that represents the 11.025kHz tone compared with the Halide DAC HD (August 2012, fig.10). There are also two pairs of low-level sidebands visible, at ±713 and ±1426Hz, these harmonically related to one another but of unknown origin.
Designing a bus-powered DAC is tricky because of the limited headroom available. But AudioQuest's DragonFly manages that trick very well, especially considering its affordable price. And I have to give a personal shout-out to AQ for the fact that their logo changes color depending on the sample rate, a feature that usefully confirms that you're sending data to the DragonFly at the correct rate.John Atkinson