Sidebar 3: Measurements
I measured iFi Audio's Pro iDSD (serial no. 3002000468, a different sample from the one auditioned by HR) with my Audio Precision SYS2722 system (see the January 2008 As We See It"). I used both the Audio Precision's S/PDIF outputs and USB data sourced from my MacBook Pro running on battery power and playing WAV and AIFF test-tone files. Apple's USB Prober utility identified the iDSD as "iFi (by AMR) HD USB Audio" from "iFi (by AMR)," and confirmed that its USB port operated in the optimal isochronous asynchronous mode. Apple's AudioMIDI utility revealed that, via USB, the iFi Pro iDSD accepted 16- and 24-bit integer data sampled at all rates up to 768kHz; its TosLink inputs accepted PCM datastreams with sample rates up to 96kHz, and its coaxial S/PDIF and AES/EBU inputs accepted data sampled at up to 192kHz.
The iFi Pro iDSD offers a choice of four output levels: HiFi Fixed and Variable and Pro Fixed and Variable. (With the volume control set to its maximum, the output levels from the Variable and Fixed settings were the same.) In solid-state output mode, the maximum output level at 1kHz from the balanced XLR jacks was 8.73V in HiFi mode, 4.32V in Pro mode, the opposite of what I was expecting from the manual. The levels from the unbalanced RCA jacks were half these figures, as expected. The maximum level from the single-ended headphone output depended on the headphone gain setting: 587mV, 2.16V, or 4.37V. Inserting the tubed stage in both of its modes reduced the maximum output level from 8.73 to 7.52V, a difference of an audible –1.3dB. The output impedance was not affected by the tube stage; in both solid-state and the two tube modes, it was a low 144 ohms from the balanced jacks, 72 ohms from the unbalanced jacks, and a very low 2.75 ohms from the headphone output. All the outputs preserved absolute polarity (ie, were non-inverting).
The Pro iDSD's impulse response depended on which of its reconstruction filters had been selected. With 44.1kHz data, the Transient Aligned filter's impulse response revealed it to be a conventional linear-phase type, with time-symmetrical ringing (fig.1). The Apodizing filter was a minimum-phase type, with all of the ringing following the single high sample (fig.2). The Bit Perfect filter in both Direct and PCM Upsampling modes had a perfect transient response (fig.3; the small amount of linear-phase ringing in this graph is due to the Audio Precision's A/D converter operating at a 200kHz sample rate). Bit Perfect+ had some slight Nyquist frequency ringing, with more after the single high sample than before (not shown). The Gibbs Transient Optimized filter had a short, minimum-phase impulse (fig.4).
Fig.9 shows the iFi's Transient Aligned frequency response at sample rates of 44.1, 96, 192, and 384kHz. The response rolls off relatively quickly above each Nyquist frequency (half the sample rate). The Gibbs filter behaved similarly, while the Apodizing filter offered a faster rolloff with 44.1kHz data (fig.10, gray and green traces)—not what I was expecting, given the results of the white-noise test. It was difficult to measure the frequency response at 44.1 and 96kHz in the Bit Perfect and Bit Perfect+ modes, as the output levels with tones above 10kHz were affected by high levels of aliased image energy.
The Pro iDSD's channel separation (not shown) was very good in all three output modes, at >120dB below 200Hz, and still 77dB L–R and 83dB R–L at 20kHz. The low-frequency noise floor was commendably free from power-supply–related spuriae (fig.11), though in Tube mode (cyan and magenta traces) some very low-level spuriae appeared at 100Hz and its harmonics. (Could this have been related to the switch-mode power supply?)
As suggested by the spectra in fig.5, harmonic distortion was not as low I would have expected. With a full-scale 50Hz tone in Solid-State, HiFi mode, the third harmonic was the highest in level at –60dB (0.1%, fig.15), though reducing the level to –3dBFS dropped the level of this harmonic to –67dB (fig.16). The second harmonic was lower in level, at –70dB in the left channel (blue trace) and –80dB in the right (red). Interestingly, though HR conjectured that he preferred the minimum-feedback Tube+ output mode because it added second-harmonic distortion, the third harmonic is still the highest in level (fig.17), though it is about 8dB higher than in the Solid-State mode. Commendably, the Pro iDSD had no problem driving the punishing 600 ohm load, with no significant increase in harmonic distortion.
Footnote 1: From "Barangrill," from For the Roses.
Fig.1 iFi Pro iDSD, Transient Aligned filter, impulse response (one sample at 0dBFS, 44.1kHz sampling, 4ms time window).
Fig.2 iFi Pro iDSD, Apodizing filter, impulse response (one sample at 0dBFS, 44.1kHz sampling, 4ms time window).
Fig.3 iFi Pro iDSD, Bit Perfect filter, impulse response (one sample at 0dBFS, 44.1kHz sampling, 4ms time window).
Fig.4 iFi Pro iDSD, Gibbs Transient Optimized filter, impulse response (one sample at 0dBFS, 44.1kHz sampling, 4ms time window).
With white noise sampled at 44.1kHz, the Transient Aligned filter rolled off the output above 20kHz (fig.5, magenta and red traces), and the aliased image of a 19.1kHz tone at 0dBFS (cyan, blue) was suppressed by >100dB. However, some harmonics of this tone are visible in this graph, the highest in level, the third, lying at –54dB (0.2%). Despite its name, the Apodizing filter appeared to offer a slower rolloff above the audioband, with the 25kHz image suppressed by 75dB (fig.6). The Gibbs Transient Optimized filter (fig.7) offered a slow rolloff with just 20dB attenuation of the image at 25kHz, and a large number of other aliased images are present. The upsampled Bit Perfect and Bit Perfect+ modes puzzled me: while the ultrasonic rolloff was extremely slow, with the expected nulls at 44.1 and 88.2kHz (fig.8, red and magenta traces), the noise floor with the full-scale 19.1kHz tone rose to the 14-bit level above 10kHz. (For what it's worth, when it was turned on the Pro iDSD blocked FM reception on the portable radio in the test lab.)
Fig.5 iFi Pro iDSD, Transient Aligned filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with data sampled at 44.1kHz (20dB/vertical div.).
Fig.6 iFi Pro iDSD, Apodizing filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with data sampled at 44.1kHz (20dB/vertical div.).
Fig.7 iFi Pro iDSD, Gibbs Transient Optimized filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with data sampled at 44.1kHz (20dB/vertical div.).
Fig.8 iFi Pro iDSD, Bit Perfect filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan), with data sampled at 44.1kHz (20dB/vertical div.).
Fig.9 iFi Pro iDSD, Transient Aligned filter, frequency response at –12dBFS into 100k ohms with data sampled at: 44.1kHz (left channel gray, right green), 96kHz (left cyan, right magenta), 192kHz (left blue, right red) (1dB/vertical div.).
Fig.10 iFi Pro iDSD, Apodizing filter, frequency response at –12dBFS into 100k ohms with data sampled at: 44.1kHz (left channel gray, right green), 96kHz (left cyan, right magenta), 192kHz (left blue, right red) (1dB/vertical div.).
Fig.11 iFi Pro iDSD, spectrum (0Hz–1kHz) of dithered 1kHz tone at 0dBFS in Solid-State mode (left channel blue, right red) and Tube mode (left cyan, right magenta) (20dB/vertical div.).
Increasing the bit depth from 16 to 24 with a dithered 1kHz tone at –90dBFS lowered the noise floor by 12dB or so (fig.12), though the 24-bit noise floor (blue and red traces) had a large number of low-level spuriae present. This behavior was not affected by the reconstruction filter used, though in Bit Perfect mode the random noise floor began to rise in the treble. With undithered data representing a tone at exactly –90.31dBFS (fig.13), the three DC voltage levels described by the data were well resolved, with a symmetrical waveform and no DC offset. With undithered 24-bit data, the result was a clean sinewave (fig.14).
Fig.12 iFi Pro iDSD, DSD512 upsampling, 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.13 iFi Pro iDSD, Gibbs Transient Optimized filter, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit TosLink data (left channel blue, right red).
Fig.14 iFi Pro iDSD, Gibbs Transient Optimized filter, waveform of undithered 1kHz sinewave at –90.31dBFS, 24-bit TosLink data (left channel blue, right red).
Fig.15 iFi Pro iDSD, balanced Solid-State output, spectrum of 50Hz sinewave, DC–1kHz, at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.16 iFi Pro iDSD, balanced Solid-State output, spectrum of 50Hz sinewave, DC–1kHz, at –3dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.17 iFi Pro iDSD, balanced Tube+ output, spectrum of 50Hz sinewave, DC–1kHz, at –3dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
With the Transient Aligned and Apodizing filters, the iDSD dealt relatively well with a full-scale mix of tones at 19 and 20kHz. Not many high-order intermodulation products were produced, and the second-order difference product at 1kHz lay at a respectably low –73dB, left channel, and –80dB, right (fig.18). As with the harmonic-distortion tests, these results were not significantly affected by upsampling to DSD512. However, as expected from its relatively poor ultrasonic rejection, the Gibbs Transient Optimized filter's behavior in this test resulted in the aliased images of the fundamental tones lying at quite high levels, and some higher-order aliased images appeared in the audioband, even though I'd reduced the signal level by 3dB (fig.19). The upsampling Bit Perfect filter behaved similarly (fig.20), though the rise in the high-frequency noise floor is again apparent.
Fig.18 iFi Pro iDSD, Transient Aligned filter, HF intermodulation spectrum, DC–30kHz, 19+20kHz at 0dBFS into 100k ohms, 44.1kHz data (left channel blue, right red; linear frequency scale).
Fig.19 iFi Pro iDSD, Gibbs Transient Optimized filter, HF intermodulation spectrum, DC–30kHz, 19+20kHz at –3dBFS into 100k ohms, 44.1kHz data (left channel blue, right red; linear frequency scale).
Fig.20 iFi Pro iDSD, Bit Perfect filter, HF intermodulation spectrum, DC–30kHz, 19+20kHz at –3dBFS into 100k ohms, 44.1kHz data (left channel blue, right red; linear frequency scale).
When I tested the iFi Pro iDSD for its rejection of word-clock jitter using undithered 16-bit J-Test data, the odd-order harmonics of the low-frequency, LSB-level squarewave were all at the correct levels (fig.21, sloping green line). However, many low-level spurious tones were present in the spectrum, and the spectral spike that represents the high-level tone at one-fourth the sample rate was slightly broadened at its base, suggesting the presence of some random low-frequency jitter. This behavior was identical when I repeated the test using the USB input (fig.22).
Fig.21 iFi Pro iDSD, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz: 16-bit AES/EBU data (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, Ò3.5kHz.
Fig.22 iFi Pro iDSD, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz: 16-bit USB data (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, Ò3.5kHz.
With so many output options and operating modes, it's easy to become confused about the iFi Pro iDSD's performance—as Joni Mitchell sang, "the crazy you get from so much choice." (footnote 1) I remain puzzled by the rise in the high-frequency noise floor in the Bit Perfect and Bit Perfect+ modes, and the low-level spuriae in the Pro iDSD's output. But there is still much to admire in its measured performance.—John Atkinson
Footnote 1: From "Barangrill," from For the Roses.
