iFi Audio iDAC & iUSBPower USB D/A processor & outboard power supply Measurements
I assessed the iFi Audio iDAC'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"); I also used my vintage Audio Precision System One Dual Domain for some tests. I used WAV files sourced from my MacBook Pro for the testing, and powered the iDAC from the iUSBPower supply. The ground switch was left in the connected position, which gave the lowest noise level with my setup. The Macintosh USB Prober utility identified the iDAC as "AMR USB Audio 2.0," and confirmed that it would operate in isochronous asynchronous mode with data with a 24-bit word length and sampled at 44.1, 48, 88.2, 96, 176.4, and 192kHz. Curiously, both USB Prober and the AudioMIDI Setup utility identified the iDAC as having a microphone input. Presumably this is available on the chip used by AMR but not implemented in the hardware.
The maximum output level from the first sample of the iFi's line outputs was 1.975V. It was the specified 3.3V from the headphone jack, this reached with a 1kHz signal at 0dBFS when the volume control was set to 2:00. Higher settings of the volume control clipped the waveform with full-scale signals. Both sets of outputs preserved absolute polarity. The output impedance was 160 ohms from the RCA jacks, around 1 ohm from the headphone jack. (Both figures include the series impedance of 6' of interconnect.)
With 44.1kHz data, the time-symmetrical ringing in the iDAC's impulse response (fig.1) indicates that the digital reconstruction filter is a conventional linear-phase type. Fig.2 is a wideband spectrum showing both the iDAC's response with 44.1kHz-sampled white noise (blue trace) and how it handles a full-scale tone at 19.1kHz (red). This graph was taken with the first sample of the iDAC; the rolloff slowed above the Nyquist frequency (half the sample rate) for the final 10dB of attenuation. With the second sample (fig.3), the digital filter rolls off very quickly, as expected from its impulse response. With both samples, there was just a trace of the aliasing product at 25kHz; the only other ultrasonic components present were the harmonics of the 19.1kHz tone. The second sample was more linear than the first, the third harmonic, the highest in level, lying at 79dB (0.01%) rather than 69dB (0.03%).
Fig.4 shows the iDAC's frequency response with data sampled at 44.1kHz (green and gray traces), 96kHz (cyan and magenta), and 192kHz (blue and red). At the two lower sample rates, the response follows the same shape as at 192kHz, though dropping off sharply just below each Nyquist frequency. Note the excellent channel matching in this graph. Channel separation (not shown) was excellent at almost 110dB RL, but 10dB less good in the other direction.
For consistency with my tests of digital products going back to 1989, I estimate a DAC's resolution by feeding it dithered data representing a 1kHz tone at 90dBFS with 16- and 24-bit word lengths and sweeping a 1/3-octave bandpass filter down from 20kHz to 20Hz. The result is shown in fig.5, with the 16-bit spectrum the top pair of traces, the 24-bit spectrum the middle pair. The increase in bit depth drops the noise floor by up to 18dB in the treble, suggesting that the iDAC has a resolution of between 18 and 19 bits, though this is limited by the rise in the noise level in the midrange and bass. The noise floor with 24-bit data also looks rather granular at high frequencies, and what is very strange is that, with 24-bit data representing a dithered 1kHz tone at 120dBFS (bottom pair of traces), the spectral peak lies at 800Hz rather than 1kHz.
Fig.5 has limited resolution, of course, so I repeated the analysis with the 90dBFS tone using a modern high-resolution FFT technique (fig.6). While the peak representing the 1kHz tone still lies at 90dBFS, the noise floor with 24-bit data is marred by enharmonic peaks; ie, their frequencies are not mathematically related to the frequency of the signal. Repeating this analysis with a 24-bit tone at 120dBFS (not shown) revealed that while there was a spectral component at 1kHz, its level was actually 125dBFS; and there was a stronger tone present at 802Hz, lying at 113dBFS. It's fair to note that this odd behavior lies at too low a level to have any audible effectsthe noise floor of typical recordings will be much higher than these artifactsbut it suggests that there is something funky in the iDAC's handling of low-level 24-bit data. Even so, with undithered 16-bit data, the iDAC clearly resolves the three DC voltage levels described by a tone at exactly 90.31dBFS, though these levels are overlaid with more noise than with the best conventional DACs (fig.7).
At low frequencies and high levels into high impedances from the RCA jacks, the iDAC's distortion signature comprised the second and fourth harmonics (fig.8), both of which are harmonically consonant even at higher levels than this. However, the first sample's left channel (blue trace) is slightly less linear than the right (red), with some very low-level higher-order harmonics visible. The second sample was fine in this respect. At higher frequencies the third and fifth harmonics begin to make an appearance with both samples, and with a punishing 600 ohm load (not shown), the iDAC's output was marred by a picket fence of distortion harmonics. While the headphone output will drive low impedances with aplomb, the RCA jacks should be used with loads of 5k ohms or higher. Into high impedances, intermodulation distortion was low (fig.9), especially with the second sample (fig.10).
An asynchronous USB connection is, in theory, free from interface jitter. While the spectrum of the iDAC's analog output while it handled 16-bit data representing the Miller-Dunn J-Test signal (fig.11) indicated that there were no jitter-related sidebands, and that the residual odd-order harmonics of the LSB-level low-frequency squarewave were not being accentuated, the spike that represents the high-level tone at 1/4 the sample rate was widened at its base. This hints at the presence of some low-level, low-frequency random jitter, and the noise floor looks a little hashier than I was expecting. With a 24-bit version of the J-Test signal (fig.12), which should consist of the central peak and a clean noise floor, there were some enharmonic tones as well as pairs of sidebands of unknown origin at ±721Hz and ±3.33kHz. Again, this suggests that something is not quite right with the iDAC's handling of 24-bit data.
In most respects the iFi iDAC measured well, especially considering its affordable price when powered directly from the computer rather than being used with the iUSBPower. And as the second sample did offer lower distortion than the first, it does look as if our original review sample was not performing optimally.John Atkinson