Devialet D-Premier D/A integrated amplifier Measurements
Sidebar 3: Measurements
I measured the Devialet D-Premier 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 measured the later, black-finish sample, serial no.00965, the mirror-finish no.00061 having been sent off for photography. Before testing an amplifier with a conventional class-B or -AB output stage, I precondition it by running it at one-third power for an hour, which thermally stresses the amplifier to the maximum extent. In theory, this test is irrelevant with the Devialet, but I did so anyway. At the end of the hour, the left side of the top panel was hot to the touch, at 131.4°F (55.3°C), the right side a little cooler, at 114.3°F (45.8°C). It might be efficient at converting wall current into speaker-driving power, but the D-Premier still runs hot. As Devialet warns in the D-Premier's manual, the amplifier needs to be well ventilated.
I didn't test the phono input's performance, as I didn't use it during my auditioning for this review. The phono stage's sound and measured performance will be covered in a Follow-Up.
The rotary volume control on the Devialet's remote control covers a range of "97.5dB" to "+30.0dB" in accurate 0.5dB steps. With the control set to "+30.0," the voltage gain into 8 ohms for an analog source and a line input was a high 52.3dB. At "0.0dB," the gain, of course, was 22.3dB. The analog inputs preserved absolute polarity (ie, were non-inverting), and the analog input impedance was moderately low, at 14k ohms at low and middle frequencies, dropping slightly at 20kHz to 12.5k ohms.
Devialet specifies the D-Premier's output impedance as <0.001 ohm. I measured 0.04 ohm at all audio frequencies, but thus includes the impedance of the 6' of speaker cable I used for the test. But the fact that the D-Premier has an extraordinarily low output impedance can be seen in fig.1, where the voltage drop as the load changes from 8 to 4 ohms, and the variation in response with our standard simulated loudspeaker, are each less than 0.1dB. Even with a 2 ohm load (green trace), the level drops by less than 0.2dB in the audioband. With the D-Premier in its standard configuration, pressing the middle bottom button on the remote rolls off the low frequencies by 3dB at 20Hz (not shown).
Fig.1 was taken with the A/D converter that digitizes the analog inputs set to the default of 96kHz sampling. As a result, the response can be seen to drop like a stone above 40kHz, and a 10kHz squarewave has the distinctively rounded shape that is due to the absence of all harmonics above the third (fig.2). The 1kHz squarewave (not shown) has the usual time-symmetrical ringing visible preceding the start of each change in level (not shown).
Channel separation via the analog inputs was excellent, at >100dB in both directions below 2.5kHz (not shown). The D-Premier is specified as having a superb signal/noise ratio of 130dB, but without conditions or reference level given. My standard test with an integrated amplifier is to short-circuit the input but set the volume control to its maximum, which is very much the worst-case condition (some have called it unrealistic). With its very high gain, the D-Premier is put at a disadvantage under these conditions. I measured the wideband, unweighted S/N, ref. 1W into 8 ohms, to be a fairly low 47.1dB in the left channel and 48.5dB in the right. Restricting the measurement bandwidth to the audioband improved these figures to 51.6 and 52.3dB, respectively, while switching in an A-weighting filter gave further improvement, to 54.9 and 55.2dB. Reducing the level of the volume control will improve these ratios, of course, and fig.3 shows the low-frequency spectrum of the D-Premier's output while it reproduced a 1kHz tone at 100W into 8 ohms with the volume control set to "0.0dB." The noise floor is a little higher in the left channel than in the right, and rises by 10dB as the frequency decreases. The only other thing to note in this graph is the presence of a power-supplyrelated tone at 120Hz, but at 110dB ref. the level of the 1kHz tone, this is inconsequential.
The Devialet is specified as having a maximum output power of 240Wpc into, I understand, 6 ohms (22dBW). Fig.4 shows how the THD+noise percentage changes with output power into 8 ohms.1 The downward slope of the trace below the "knee" indicates that what distortion is present is actually buried in the noise floor. We define clipping as the power when the THD reaches 1%; fig.4 shows that the D-Premier was putting out 180Wpc into 8 ohms (22.55dBW) at 1% THD. Fig.5 shows the behavior into 4 ohms. The trace is similar in shape to that into 8 ohms, except that, very unusually, it slopes backward above the "knee," which occurs at 245.8Wpc into 4 ohms (20.9dBW). What is happening is that, with sustained drive at very high power, either the power supply starts to collapse or the DSP protection starts to operate, both reducing the power delivery. Because of this behavior, I didn't test the maximum power into 2 ohms.
I plotted how the output power changes with frequency at a level, 20V, equivalent to 50W into 8 ohms and 100W into 4 ohms, where the power supply would remain stiff. The results are shown in fig.6. Below 10kHz, the measurement is dominated by noise; above that frequency, and unlike a conventional class-D amplifier, there is almost no rise in THD. However, I haven't shown the behavior into 2 ohms as, again, the power-supply voltage would start to decrease after a few seconds' operation at 20V.
The lower trace in fig.7, taken at a level just below the discontinuity in fig.4, confirms that actual distortion in the D-Premier's output is buried beneath the noise. However, FFT analysis reveals there to be harmonics present at high powers, albeit at a very low level (fig.8). The two highest in level, the second and third harmonics, lie at 106dB (0.0005%) and 104dB (0.0006%), respectively, with all other harmonics lying at or below 110dB (0.0003%). Conventional class-D amplifiers tend to perform poorly when fed high-frequency intermodulation. By contrast, the D-Premier, given a mix of 19 and 20kHz tones at a peak level of 20W into 4 ohms (the highest level the amplifier would deliver with this signal without the power supply collapsing), performed well on this test (fig.9). The 1kHz difference tone lay at almost 110dB, and the higher-order intermodulation products were all below 90dB (0.003%).
Turning to the D-Premier's digital inputs, I measured its performance at the speaker outputs, as that is how I used the amplifier. With the volume control set to "0.0dB," a full-scale 1kHz tone resulted in a level of 30.79V into 8 ohms, equivalent to a power of 118.5Wpc, just under 2dB below clipping into this load. This suggests that, with digital inputs, the volume control should be kept below "+1.5dB." The digital inputs were again non-inverting. Although the S/PDIF and AES/EBU inputs locked on to datastreams with sample rates from 32 to 192kHz, including 88.2 and 176.4kHz, the "AIR WiFi input was restricted to 96kHz and below.
Fig.10 shows the D-Premier's frequency response for its digital inputs with data sampled at 44.1, 96, and 192kHz. Unusually, the 192kHz rate gives only a slight increase in bandwidth compared with 96kHz. Digital channel separation (not shown) was similar to that for analog inputs. To remain consistent with the measurements of DAC resolution I have performed since 1989, I used a swept-bandpass technique to generate the traces in fig.11, which represent a dithered tone at 90dBFS with 16- and 24-bit data. Repeating the analysis with a modern FFT technique gave a similar picture (fig.12), confirming that the only power-supplyrelated spuriae present was a small amount of 120Hz, at 110dB. The increase in bit depth from 16 to 24 with both these graphs dropped the noise floor by around 12dB, implying a resolution of 18 bits or so, which is good, if not quite up to the standard set by the best-measuring standalone processors, such as the Bricasti M1, MSB Diamond DAC IV, NAD M51, or Weiss DAC202.
There is some peculiar scalloping of the noise floor visible with 24-bit data in the right channel in fig.12 (red trace). I am not sure what this means. Repeating the analysis with 16- and 24-bit data over WiFi using the D-Premier's AIR input and the Devialet iTunes streamer gave an anomalous result (fig.13). With 16-bit data, the 1kHz tone lay at the correct 90dBFS in the left channel (cyan trace), the noise floor lay at the correct level, and no harmonics were visible. However, the right channel's noise floor (magenta trace) lay about 5dB higher in level. Even more peculiarly, sending 24-bit data gave only a slight lowering of the noise floor in the left channel (blue trace), but a raising of the floor in the right channel (red). I tried rebooting the D-Premier by unplugging it from the wall and then powering it back up, but this did not restore proper operation with 24-bit data. This was with v.1.4.0 of Devialet's AIR client; I imagine it is something that can be fixed in a future software or firmware release.
With S/PDIF and AES/EBU data, the D-Premier's reproduction of an undithered sinewave at exactly 90.31dBFS was essentially perfect (fig.14), with undithered 24-bit data giving a well-defined sinewave (fig.15).
Feeding the D-Premier a 24-bit version of the 44.1kHz Miller-Dunn J-Test via one of the TosLink inputs gave the spectrum shown in fig.16. No data-related jitter components are visible. However, a pair of sidebands around the 11.025kHz tone at ±120Hz can be seen. Although these lie below 120dBFS (0.0001%) and are presumably irrelevant to sound quality, they perhaps result from some power-supply interference at the DAC's voltage-reference pin. Repeating the test with Devialet AIR WiFi data gave rise to the same 120Hz sidebands (fig.17), but now some broadening of the central peak can be seen. And as with fig.13, the left channel's noise floor (blue trace) is not much lower than the 16-bit floor, while the right channel's floor is significantly higher in level (red). I suspect that this loss of resolution via WiFi is the reason I ended up preferring using a standalone USB-S/PDIF converter for the bulk of my auditioning of the D-Premier.
Other than its slightly anomalous behavior via WiFi, the Devialet D-Premier's measured performance is a testament to what can be obtained with switching-amplifier technology. However, sustained high-power operation into speakers with an impedance of 4 ohms or below is best avoided. John Atkinson
Footnote 1: Because the D-Premier's output stage includes a class-D element, I performed all the distortion measurements using Audio Precision's AUX-0025 high-power, passive low-pass filter between the amplifier's output and the SYS-2722's input. But this really didn't appear to be necessary; unlike with the class-D amplifiers we reviewed in December, I could detect no ultrasonic switching noise in the D-Premier's output.John Atkinson