Digital Processor Reviews

I examined the PS Audio PerfectWave DirectStream's electrical performance with my Audio Precision SYS2722 system (see www.ap.com and the January 2008 "As We See It"). To test its USB input, I used my 2012 MacBook Pro running on battery power. Unless stated otherwise, all measurements were performed from the processor's balanced outputs. I don't yet have files of DSD-encoded test signals, so I assessed the DirectStream's measured performance with PCM-encoded files, with sample rates extending from 44.1 up to 352.8kHz. All testing was done with the DirectStream running on the newer v.1.1.4 firmware and the volume control was set to its maximum.

Apple's USB Prober utility identified the PS Audio as having the Product String "PS Audio USB Audio 2.0" and the Manufacturer String "PS Audio," and confirmed that the DirectStream processor operates in the optimal isochronous asynchronous mode. The USB input worked with data of sample rates from 44.1 to 352.8kHz, while the AES/EBU and S/PDIF inputs successfully locked to datastreams with sample rates up to 192kHz; as usual, the TosLink input was restricted to a maximum rate of 96kHz. The maximum output voltage at 1kHz was 2.85V balanced and 1.44V unbalanced. Switching in the output attenuator with the touchscreen reduced these levels by just over 20dB, to 273 and 106mV, respectively. Both outputs preserved absolute polarity (ie, were non-inverting). At low and midrange frequencies, the DirectStream's output impedance was a fairly low 296 ohms balanced and 124 ohms unbalanced. Both impedances rose slightly at the top of the audioband, to a respective 313 and 145 ohms, but this will be inconsequential.

The PS Audio's impulse response with 44.1kHz data was a standard linear-phase type, with symmetrical "ringing" before and after the single sample at 0dBFS (fig.1). Fig.2 shows the results of a test suggested to me by Jürgen Reis of MBL, which characterizes the behavior of a digital device's reconstruction filter with FFT-derived spectral analysis up to 100kHz. The device under test decodes first 44.1kHz data representing white noise (magenta and red traces), then 44.1kHz data representing a full-scale tone at 19.1kHz (cyan, blue). With the tone, the second, third, and fourth harmonics, respectively at 38.2, 57.3, and 76.4kHz, can be seen, albeit at respectably low levels. But there is no trace of the aliasing product at 25kHz, this being completely suppressed by the reconstruction filter. The noise trace shows that the filter rolls off the processor's output sharply above 20kHz, and reaches the noise floor before the Nyquist frequency (half the sample rate; indicated by the vertical green line). However, with both noise and tone data, the noise floor rises at ultrasonic frequencies, this presumably due to the PS Audio's DSD upsampling. Peculiarly, the two channels behave differently, the rise in the left channel's floor (cyan and magenta traces) starting at 40kHz, while the right channel's floor (blue, red) doesn't start to rise until 60kHz.

Fig.3 displays a more conventional representation of the DirectStream's frequency response, taken with data sampled at 44.1, 96, and 192kHz. In each case, the audioband response has a very slight rolloff in the low bass, reaching –0.8dB at 10Hz, but is otherwise flat up to 20kHz. The reconstruction filter rolls off the output just below the Nyquist frequency with the two lower sample rates, but the 192kHz response (blue and red traces) rolls off gently above 50kHz, reaching –3dB at 60kHz and –9dB at 80kHz. I also plotted the response with data sampled at 352.8kHz but haven't shown it, as it was identical to the 192kHz behavior. Fig.3 was taken from the DirectStream's balanced outputs. The cyan and magenta traces in fig.4 show the balanced output response at 96kHz, but plotted with a higher-resolution vertical scale and offset by –1dB for clarity. The blue and red traces in this graph were taken from the unbalanced output; it is slightly different from the balanced behavior in having a very slight rise in level above 12kHz, reaching +0.1dB at 25kHz. I doubt that this will have any audible consequences, but it is unusual.

Channel separation (not shown) was very good, at 95dB or greater across the audioband. Spectral analysis of the low-frequency noise floor while the PS Audio decoded 24-bit data representing a 1kHz tone at 0dBFS (fig.5) revealed that power-supply spuriae were extremely low in level, with the full-wave–rectified product at 120Hz present only in the right channel (red trace), but even so, at –114dB, this will be inconsequential. The DirectStream's circuit-board layout must have been done with a careful eye on grounding issues.

Things didn't look so good, however, when I performed a wider-band spectral analysis with the PS Audio processing dithered data representing a 1kHz tone at –90dBFS with 16-bit (fig.6, cyan and magenta traces) and 24-bit data (blue, red). With 16-bit data, the noise floor is dominated by the dither used to encode the signal, though a trace of second-harmonic distortion is visible. But when the bit depth was increased to 24, which was correctly indicated on the front-panel display, the noise floor dropped at most by 5dB, suggesting that the DirectStream DAC has only about 17 bits of resolution. This graph was taken with AES/EBU data; I got the same result with USB data, and though fig.6 suggests that the PS Audio DAC should just be able to resolve a 24-bit tone at –120dBFS, spectral analysis showed that it couldn't (figs.7 and 8). Its higher-than-expected noise floor meant that the PS Audio couldn't correctly resolve the waveform of an undithered 16-bit tone at exactly –90.31dBFS (fig.9). (For superb behavior on this test, see fig.6 in the review of the Antelope Zodiac Platinum elsewhere in this issue.)

While distortion on higher-frequency tones was acceptably low, even into 600 ohms—with a full-scale 1kHz tone, the second and third harmonics lay at –77dB (0.014%) and –71dB (0.029%), respectively—the DirectStream was less linear at low frequencies. Fig.10 was taken with a 50Hz tone into the kind load of 100k ohms. Even so, the second harmonic lay at –43dB (0.7%) and the third at –52dB (0.27%), with higher harmonics also visible. Dropping the signal level by 10dB (fig.11) reduced the level of the second harmonic by about 9dB with respect to the level of the fundamental tone, but the third harmonic dropped by just 1dB or so. At least this behavior didn't worsen into 600 ohms (not shown), and the DirectStream behaved very well when it came to high-frequency intermodulation distortion (fig.12). All the intermodulation products lay at –80dB (0.01%) and below, though some low-level spurious tones can be seen. These spuriae remained visible when I switched into circuit a brick-wall low-pass filter at 40kHz.

I obtained anomalous results when I tested the PS Audio DirectStream's rejection of wordclock jitter. Fig.13 shows a narrowband analysis of the processor's output while it decoded 16-bit J-Test data. While the central spike that represents the high-level tone at 11.025kHz is superbly well defined, the noise floor is closer to the 15- rather than the 16-bit level. In addition, while the odd-order harmonics of the low-frequency, LSB-level squarewave decay correctly with increasing frequency, they are all higher than their correct level (green line). With 24-bit J-Test data (fig.14), the noise floor remains too high in level, though no harmonics can now be seen.

It is important to note that figs.13 and 14 are free from jitter-related artifacts; these graphs were taken with AES/EBU data. I got identical results with TosLink- and USB-sourced data—but something strange is going on.

In many ways, PS Audio's DirectStream DAC measures superbly well. But I was somewhat bothered by its ultimate lack of resolution with data capable of higher-than-CD resolution, which I suspect lay behind AD's finding the processor to sound "a bit rounded off" and lacking in immediacy. Whether or not that will be an issue will depend on the listener's taste and the character of his or her other components. It is also fair to point out that the DirectStream's introduction of random noise at the 17-bit level will be sonically preferable to a processor whose errors consisted of enharmonic tones. But I was bothered by the PS Audio's poor linearity at low frequencies. Yes, some extra second- and third-harmonic content at low frequencies will add some "phatness" to the sounds of bass instruments; and as the nonlinear transfer function has been significantly improved in the midrange and above, that "phat" quality will not be accompanied by the high-frequency harshness of intermodulation. But the engineer in me doesn't like to see it.—John Atkinson