NAD M2 Direct Digital integrated amplifier Measurements
The NAD M2 is not the first power D/A converter to have crossed my test bench: the Sharp SM-SX100, reviewed by Michael Fremer in July 2000 had that honor. But as I wrote at the time in the measurements section that accompanied that review, "On the face of it, an amplifier accepting digital input data and operating entirely within the digital domain is a very attractive idea. But as Sharp's SM-SX100 reveals, it takes heroic engineering to make it work, and there are still some compromises involved, particularly in achieving sufficient dynamic range."
The question facing the NAD M2, therefore, was whether it had implemented the concept without compromise. But first, as class-D amplifiers have significant levels of ultrasonic switching noise present at their speaker terminals, even with the obligatory low-pass filter in series with their output, accurately assessing their performance is not trivial. This ultrasonic noise can drive the input stage of an analyzer into slew-rate limiting, leading to inaccurate distortion measurements. See Bruce Hofer's PowerPoint presentation on this subject, which can be downloaded here. I therefore performed the distortion, noise, and channel-separation tests on the M2 using the Audio Precision AUX-0025 passive filter discussed by Hofer.
I measured the M2's performance with Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see the January 2008 "As We See It" and www.ap.com); for some tests, I also used my vintage Audio Precision System One Dual Domain. Before I test an amplifier, I run it for 60 minutes at one-third its specified power into 8 ohms. Thermally, this is the worst case for an amplifier with a class-B or -AB output stage, but for a class-D amplifier such as the M2 it represents nothing out of the ordinary. Even so, at the end of the hour, the M2's chassis was comfortably warm.
Looking first at the analog inputs, both the balanced and unbalanced inputs offered a maximum gain (with the volume control set to "10.0dB") of 39.55dB into 8 ohms. Both inverted signal polarity with the Polarity set to Positive. Without any Input Offset, the analog inputs overloaded at 2.025V RMS, which is a bit too close for comfort to the maximum output of many CD players. I recommend adding a little headroom by trimming the analog input's gain by 3dB with the M2's Menu buttonwhich, I'm told, is how production samples are being shipped. The unbalanced analog input impedance was close to specification, at 33.2k ohms at low and middle frequencies, dropping lightly to 29.2k ohms at 20kHz. The balanced input impedance was twice these values.
The M2's output impedance didn't appear to vary with the loudspeaker-impedance compensation setting used. It was extremely low at low and middle frequencies, at 0.060.07 ohm, which is close to the residual impedance of the 6' speaker leads I use in my testing. At 20kHz, however, it varied between 1 and 1.5 ohms, depending on the load impedance I used to take the measurement.
The speaker-impedance compensation setting did have a major effect on the amplifier's frequency response, however. Fig.1 shows the response into 8, 4, and 2 ohms, and into our standard simulated loudspeaker, with the compensation set to ">8 ohms." (The digitizer sample rate was set to 192kHz for these measurements.) Despite the very low impedance, the variations into our standard loudspeaker load reach ±0.4dB in the treble, and the top octave can be seen to roll off into the lower impedances, with the 2 ohm response (green trace) down 6dB at 20kHz. Switching the compensation to "4 ohms" gives the family of traces in fig.2. The 2 ohm response is now 3dB at 20kHz, the 4 ohm output is flat to 40kHz (cyan and magenta traces), and the 8 ohm output rises to +6.5dB between 80 and 90kHz (blue and red). As you'd expect, the "2 ohm" compensation gives a flat audioband response into 2 ohms (fig.3), but with now a 12dB ultrasonic peak into 8 ohms. I recommend that you look carefully at the top-octave response of your loudspeakers, and take care not to set the M2's compensation lower than the average impedance in that region.
Fig.1 NAD M2, >8 ohms speaker compensation, frequency response at 2.83V into: simulated loudspeaker load (gray), 8 ohms (left channel blue, right red), 4 ohms (left cyan, right magenta), 2 ohms (green). (2dB/vertical div.)
Fig.2 NAD M2, 4 ohms speaker compensation, frequency response at 2.83V into: simulated loudspeaker load (gray), 8 ohms (left channel blue, right red), 4 ohms (left cyan, right magenta), 2 ohms (green). (2dB/vertical div.)
Fig.3 NAD M2, 2 ohms speaker compensation, frequency response at 2.83V into: simulated loudspeaker load (gray), 8 ohms (left channel blue, right red), 4 ohms (left cyan, right magenta), 2 ohms (green). (2dB/vertical div.)
The A/D converter used to digitize the M2's analog inputs has a flat response within the audioband, with a sharp cutoff associated with the sample rate chosen (fig.4). I recommend using a sample rate of 48kHz for CD-based analog sources, and 96kHz for phono preamps. The linear-phase digital filter associated with the A/D converter gives rise to the usual Gibbs Phenomenon symmetrical "ringing" with squarewaves: fig.5 shows a 1kHz squarewave digitized at 96kHz, and fig.6 a 10kHz squarewave digitized at 192kHz.
Fig.4 NAD M2, analog input, A/D converter frequency response at sample rates of 48kHz (left channel gray, right green), 96kHz (left cyan, right magenta), and 192kHz (left blue, right red). (1dB/vertical div.)
Fig.5 NAD M2, analog input, 96kHz sampling, small-signal 1kHz squarewave into 8 ohms.
Fig.6 NAD M2, analog input, 192kHz sampling, small-signal 10kHz squarewave into 8 ohms.
Channel separation at 1kHz via both analog and digital inputs was better than specified, at >95dB (analog) and >103dB (digital), both in both directions. The separation decreased at the top of the audioband, to the specified 80 and 90dB, respectively. Without any additional low-pass filtering on the output and with the input shorted but the volume control at its maximum, there was around 100mV of ultrasonic noise present at the speaker terminals, which was reduced to 20.6mV by the Audio Precision AUX0025 low-pass filter. The wideband, unweighted signal/noise ratio, ref 2.83V into 8 ohms, again with the analog input shorted, was thus limited to 42.7dB. This improved to a respectable 76.6dB when the measurement was restricted to the audioband, and to 84dB when A-weighted. The specified 95dBA ref. 1W into 8 ohms was probably taken with the volume control set to 0dB, therefore.
As shown in fig.7, the M2's frequency response via its digital input depended on the datastream's sample rate, of course. The input locked to data with sample rates ranging from 32 to 192kHz, and the display did show the correct sample rate when the status flag in the incoming datastream was correctly set. (If I left that flag blank, the M2 did operate at the incoming sample rate despite the display showing "44.1kHz.") A gentle rolloff above the audioband reached 1.5dB at 90kHz with 192kHz data (not shown). The channel matching can be seen to be superb. With the volume control set to "0.0dB," a 1kHz tone with a digital level of 20dBFS resulted in a level at the speaker terminals of 5.935V into 8 ohms, suggesting that, unless the volume is backed off a little, the amplifier will be driven into a couple of dBs' worth of clipping with full-scale digital transients.
Fig.7 NAD M2, digital input, frequency response at 2.83V into 8 ohms with 8 ohms speaker compensation, with sample rates of: 32kHz (left channel gray, right green), 44.1kHz (left blue, right red), 96kHz (left cyan, right magenta), 192kHz (left blue, right red). (1dB/vertical div.)
As a power amplifier capable of swinging at least 10 times the output voltage of a typical D/A converter, the M2 offered superb dynamic range. Fig.8 shows spectral analyses of the M2's output taken with a swept bandpass filter while the amplifier decoded 16-bit data representing a dithered 1kHz tone at 90dBFS, and 24-bit data representing dithered 1kHz tones at 90 and 120dBFS. (The external Audio Precision passive low-pass filter was in circuit for these measurements, along with an internal 20kHz brick-wall filter, as recommended by Audio Precision.) Not only are all the traces free from any harmonic or power-supplyrelated spuriae, the increase in bit depth lowers the noise floor by up to 24dB. FFT analysis (fig.9) shows a similar improvement, suggesting that the M2 has true 20-bit dynamic range, which is the state of the art of real-world digital decoding. There is thus plenty of resolution available to allow effective attenuation in the digital domain.
Fig.8 NAD M2, digital input, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (top) and 24-bit data (middle at 2kHz), and dithered 1kHz tone at 120dBFS with 24-bit data (bottom at 1kHz). (Right channel dashed.)
Fig.9 NAD M2, digital input, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (left channel cyan, right magenta) and 24-bit data (left blue, right red).
Linearity error with 24-bit data was negligible down to 120BFS, which, in conjunction with the superb resolution, meant that the M2's reproduction of an undithered 16-bit tone at exactly 90.31dBFS was essentially perfect (fig.10). The three DC voltage levels are clearly resolved, with superb waveform symmetry and the Gibbs Phenomenon "ringing" unobscured by noise. Increasing the word length to 24 bits produced an excellent sinewave (fig.11).
Fig.10 NAD M2, digital input, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data (left channel blue, right red).
Fig.11 NAD M2, digital input, waveform of undithered 1kHz sinewave at 90.31dBFS, 24-bit data (left channel blue, right red).
Assessing the M2's maximum power was a bit tricky because, like all NAD amplifiers, it offers more power for brief transients compared with its steady-state power delivery. Fig.12 shows the result of using a level-stepped tone to plot the THD+noise percentage (with the AP passive filter and with Soft Clipping turned off) into 8 and 4 ohms with both channels driven, and into 2 ohms with one channel driven. At clipping (defined as 1% THD+N) with a 1kHz tone, the M2 delivered 303Wpc into 8 ohms (24.8dBW), 389Wpc into 4 ohms (22.9dBW), and 469W into 2 ohms (20.7dBW). (In each case, the speaker compensation was set to the appropriate value.) However, the amplifier is clearly happiest driving loads above 2 ohms, as shown by the plot of the THD+N percentage against frequency (fig.13, which shows useful data only up to 6.7kHz, due to the 20kHz brick-wall filter used for this test). The distortion is very low at low frequencies into both 8 and 4 ohms, but is considerably higher into 2 ohms. In addition, the higher the frequency and the lower the impedance, the higher the THD+N, though in this respect the M2 is still an order of magnitude better than some other class-D amplifiers I have measured.
Fig.12 NAD M2, distortion (%) vs 1kHz continuous output power into (from bottom to top below 100W): 8, 4, 2 ohms.
Fig.13 NAD M2, THD+N (%) vs frequency at 10V into: 8 ohms (left channel blue, right red), 4 ohms (left cyan, right magenta), 2 ohms (green).
Even with the measurement system's low-pass filtering, the residual distortion at low power lay below the noise floor. I therefore looked at the residual waveform accompanying a 1kHz tone at 89W into 4 ohms, and averaged 64 traces to drop the noise that would otherwise obscure it (fig.14). The primary harmonic can be seen to be the third, though some higher-order products are present. At low frequencies into 8 ohms, the distortion is predominantly second and third harmonic (fig.15), though these spuriae are all below 112dB (0.00025%). Into lower impedances and at higher frequencies, the third and higher odd-order harmonics dominate (fig.16), though these are all still very low in level. With the decreasing linearity at high frequencies seen in fig.13, the M2's performance on the high-frequency intermodulation test was as expected (fig.17), with some high-order products present, though the second-order difference tone at 1kHz lay at a low 94dB (0.0015%). However, even the highest-level spuriae, at 18 and 21kHz, lay at 72dB (0.025%), which is fine.
Fig.14 NAD M2, 1kHz waveform at 89W into 4 ohms (top), 0.0049% THD+N; distortion and noise waveform with fundamental notched out (bottom, not to scale).
Fig.15 NAD M2, spectrum of 50Hz sinewave, DC10kHz, at 97.5W into 8 ohms (left channel blue, right red; linear frequency scale).
Fig.16 NAD M2, spectrum of 1kHz sinewave, DC10kHz, at 97.5W into 8 ohms (left channel blue, right red; linear frequency scale).
Fig.17 NAD M2, HF intermodulation spectrum, DC24kHz, 19+20kHz at 192W peak into 4 ohms (linear frequency scale).
Finally, I tested the M2's rejection of word-clock jitter by feeding a 16-bit version of the diagnostic J-Test tone from the soundcard of my test-lab PC via 15' of plastic TosLink. The resulting narrowband spectrum of the amplifier's output is shown in fig.18. The central spike representing the high-level Fs/4 tone shows very little spectral spreading at its base, and the harmonics of the Fs/192 LSB-level squarewave lie at the residual level. Other than the fact that the noise floor in the right channel (red trace) is a little higher than in the left (blue), this is state-of-the-art performance.
Fig.18 NAD M2, line output, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit data. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz (left channel blue, right red).
It is very satisfying to be able to discuss a component's measured performance without having to scratch my head over some or another idiosyncrasy. The NAD Masters Series M2 Direct Digital amplifier falls readily into that category.John Atkinson