Integrated Amp Reviews

Sidebar 3: Measurements

I measured the Naim Uniti Nova using my Audio Precision SYS2722 system (see the January 2008 "As We See It") and before I did so, I installed the Naim app on my iPad mini and used it to reset the Naim to the factory default settings. (The Uniti had no trouble logging on to my WiFi network.) Usually, before I test an amplifier, I precondition it with both channels driving a 1kHz tone at one-third power into 8 ohms for an hour. However, the Naim turned itself off after 15 minutes, displaying the message "Over temperature, please wait for the product to cool down." The top panel above the side-mounted heatsinks was indeed very hot, at 118.1°F (47.8°C). After the Uniti Nova had cooled down a bit, it turned itself back on and I continued the testing.

Looking first at the analog inputs, the maximum voltage gain at 1kHz from the speaker terminals into 8 ohms measured 35.35dB. The gain at the preamplifier outputs was 6.7dB. These inputs preserved absolute polarity (ie, were non-inverting) at all three sets of outputs. The input impedance was a usefully high 50k ohms at low and middle frequencies, dropping inconsequentially to 42k ohms at 20kHz.

The headphone output impedance was extremely low, at 0.5 ohms at all audio frequencies. The output impedance at the speaker terminals was 0.3 ohm at all audio frequencies; as a result, the modulation of the Naim's frequency response with our standard simulated loudspeaker was just ±0.2dB (fig.1, gray trace). The response is down by 2dB at 20kHz, and, peculiarly, it drops off very rapidly above 20kHz. When I looked at the Uniti Nova's reproduction of a 1kHz squarewave (fig.2), I understood what was happening. The overshoot and ringing indicate that the Naim converts its analog inputs to digital, apparently with a sampling rate of 48kHz. This means, of course, that a 10kHz squarewave will be reproduced as a sinewave (fig.3), all the odd-order harmonics that contribute to the square shape being stripped off by the anti-aliasing filter. The volume control operated in steps of approximately 0.4dB, with the unity-gain setting at "43" of a possible 100. However, as set the analog inputs overloaded at 2.65V, which means that source components with a maximum output level higher than this should be avoided.

Channel separation via the analog inputs was good rather than great, at 71dB R–L and 84dB L–R at 2kHz, decreasing by 20dB at the top of the audioband (not shown). Spectral analysis of the Naim's low-frequency noise floor (fig.4) revealed a somewhat high level of random noise, as well as AC supply components at 120 and 240Hz. (The measurements were taken with the Audio Precision's output floating and the Naim's rear-panel switch set to Grounded, the default position, which gave the lowest level of supply harmonics.) The unweighted wideband signal/noise ratio, taken with the input shorted to ground but the volume control set to its maximum, the worst case, was just 63.9dB ref. 1W into 8 ohms. This improved slightly, to 66.6dB, when the measurement bandwidth was restricted to the audioband, and to 69.3dB when A-weighted.

Figs. 5 and 6 plot the percentage of THD+noise in the Naim's output against power into 8 and 4 ohms. Specified as delivering 80Wpc into 8 ohms (19dBW) and 155Wpc into 4 ohms (18.9dBW), the Uniti Nova delivered 103Wpc into 8 ohms at clipping (1% THD+N) with both channels driven (20.1dBW), and 160Wpc into 4 ohms (19dBW). The wall voltage was a little high (123.8V) when I performed this test with the Naim clipping, which goes some way toward explaining why it delivered more than its specified power. I measured how the THD+N varied with frequency at a level, 20V, where I could be sure I was looking at distortion rather than noise (fig.7). While there was a rise in THD in the top two octaves, this was relatively small. The distortion itself comprised the relatively innocuous second and third harmonics (fig.8), and intermodulation distortion was respectably low in level (fig.9).

Turning to the Uniti Nova's behavior with digital data, I tested it using the Audio Precision's TosLink and coaxial S/PDIF inputs, and repeated the tests with WAV, AIFF, and DSD files stored on an SDcard that I plugged into the rear-panel port. The Naim successfully played PCM data sampled up to 384kHz, as well as both DSD64 and DSD128 files. With the volume control set to "100," a 1kHz tone at –20dBFS gave rise to an output of 558mV at the preamplifier outputs, 158mV from the headphone jack, and 15.23V from the speaker outputs, the latter equivalent to 29W into 8 ohms. As the amplifier clips at 28.7V, this suggests that the volume control not be set above "65" with digital sources. With all but the low-level tests, I measured the Uniti Nova's digital performance at the preamplifier outputs, but with the volume control set to avoid clipping at the speaker terminals.

The Naim's impulse response with 44.1kHz data (fig.10) indicates that the reconstruction filter is a minimum-phase type, with all the ringing following the single sample at 0dBFS. With 44.1kHz-sampled white noise (fig.11, red and magenta traces), the Uniti Nova's response didn't start to roll off until just below half the sample rate (vertical green line). An alias at 25kHz of a full-scale tone at 19.1kHz (blue and cyan traces) is therefore visible. Distortion harmonics of this tone are also visible, the second harmonic being the highest in level at –66dB (0.05%).

When I examined the Naim's digital frequency response with S/PDIF data, I got what appeared to be anomalous results (fig.12). With 44.1kHz data (gray and green traces) the output stopped just above 20kHz, but with data sampled at 96kHz (cyan, magenta) and 192kHz (blue, red) the ultrasonic response was down by 9dB at 29kHz. These responses were measured at the preamplifier outputs; repeating them at the speaker terminals gave the same result, other than the rolloff above 20kHz being slightly faster. As a check, I measured the responses again using WAV files stored on an SD card, but found no differences from S/PDIF data. The Uniti Nova's Burr-Brown DAC chip does function at sample rates up to 192kHz; I suspect that the Uniti Nova downsamples high-resolution data so that its DSP can be applied to those data.

When the Uniti decoded dithered 16- and 24-bit data representing a 1kHz tone at –90dBFS, with the volume control set to "100" to minimize the effect of its selfnoise, the increase in bit depth dropped the noise floor by around 20dB (fig.13), which implies resolution of greater than 19 bits. With an undithered 16-bit tone at exactly–90.31dBFS (fig.14) the waveform was symmetrical, but the three DC voltage levels described by the data are somewhat obscured by high-frequency noise. With undithered 24-bit data (fig.15) the result was a somewhat noisy sinewave.

As suggested by fig.11, the digital inputs' distortion signature was primarily the subjectively benign second harmonic (fig.16), and while actual intermodulation distortion was low in level (fig.17), the noise floor didn't look random. With 16-bit J-Test data, the resultant spectra (fig.18) were identical, whether I fed the Naim optical or coaxial S/PDIF data, or played the same data stored on the SDcard. A slight widening of the spectral spike that represents the high-level tone at one-quarter the sample rate can be seen, but all the odd-order harmonics of the LSB-level, low-frequency squarewave are reproduced at the correct levels (slanting green line). However, a pair of sidebands at ±120Hz can be seen in the left channel's output, these obviously related to the power supply. These sidebands are also evident with 24-bit J-Test data (fig.19), though the spectrum is otherwise superbly clean.

Other than its idiosyncratic behavior with high-sample-rate data, the Naim Uniti Nova's measured performance reveals it to be well sorted, as they say in the UK.—John Atkinson