Mark Levinson No.30 Reference Digital Processor Measurements

Sidebar 2: Measurements

Although there is very little correlation between sound quality and measurements, I expected the No.30 to have superb test-bench performance. This was indeed the case: the No.30 was the best-measuring digital processor I've tested. The following measurements were taken from the No.30's unbalanced outputs unless otherwise noted.

The No.30's output level when decoding a full-scale (0dBFS) 1kHz sinewave was 2.095V, 0.4dB higher than the CD standard of 2V. This level is, however, slightly lower than most high-end digital processors. I must reiterate the need for level matching when comparing digital processors. The level from the balanced outputs under these conditions was 4.14V (channel balance was within an astounding 0.008dB!).

Frequency response (from the balanced outputs; the unbalanced was identical) is shown in fig.1 and conforms to Madrigal's specification of 20Hz to 20kHz, +0, –0.2dB. There is exactly a 0.2dB rolloff at 20kHz, typical of most digital processors. Because de-emphasis is performed in the digital domain by the NPC digital filter (instead of in the analog domain by a passive or active filter), I expected no de-emphasis error. That's exactly what I found: the de-emphasis curve (fig.2) exactly matched the frequency-response curve. Incidentally, the owner's manual has two paragraphs on de-emphasis, taken from the Proceed products, incorrectly indicating that de-emphasis is performed in the analog output stage.

Fig.1 Mark Levinson No.30, balanced frequency response at –12dBFS into 100k ohms (right channel dashed, 0.5dB/vertical div.).

Fig.2 Mark Levinson No.30, de-emphasis error (right channel dashed, 0.5dB/vertical div.).

When I measured the No.30's channel separation, no trace appeared on the computer screen! I quickly realized that the No.30's interchannel crosstalk was so low that the standard plot would need rescaling. Typically, the crosstalk horizontal scale is between –50dB and –120dB. In the No.30's case, crosstalk was lower than –120dB at all frequencies. In fact, it was –140dB at 125Hz, an amazingly low value. Moreover, the L–R and R–L crosstalk curves virtually overlap, meaning that their performances were identical. Again, note the different scale in fig.3 when comparing to other crosstalk plots. The balanced crosstalk curves (not shown) were identical to the unbalanced curves.

Fig.3 Mark Levinson No.30, channel separation (5dB/vertical div.).

Looking at a spectral analysis of the No.30's output when decoding a –90dB, 1kHz sinewave produced the plot in fig.4. Note the very low level of noise, especially the absence of power-supply–related junk in the low frequencies. This is perhaps the best-looking converter I've measured in this respect. Also note that the 1kHz tone peaks at exactly the –90dB horizontal division, a hint that the No.30 has good low-level linearity.

Fig.4 Mark Levinson No.30, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS, 16-bit data. (Right channel dashed.)

Fig.5, the No.30's departure from linearity with a dithered signal, confirms that it indeed has excellent low-level performance. Linearity error at –90dB was +0.07dB (right channel) and +0.12dB (left), far better than Madrigal's specification of +1.7dB below –90dB. At –100dB, linearity error was only +0.39dB (L) and +0.38dB (R). Again, this is among the best low-level linearity performance measured. I suspect that the DAC is actually perfect at –100dB; the identical positive error in both channels is probably noise.

Fig.5 Mark Levinson No.30, departure from linearity, 16-bit data (right channel dashed, 2dB/vertical div.).

Fig.6 Mark Levinson No.30, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 100k ohms, balanced (linear frequency scale).

A test that has occasionally proved useful in the past in revealing D/A problems is to drive the unit under test with data representing an equal mix of 19kHz and 20kHz tones, the resulting waveform peaking at 0dB. Processors that produce much 1kHz difference product, or higher-order sidebands at 18kHz and 21kHz, can sound a little hard in the upper midrange. Looking at the spectrum of the No.30's analog output, however, revealed the processor to produce no 1kHz product above the noise floor of the FFT analyzer, with vestigial higher-order products, an excellent result.

The No.30's reproduction of a 1kHz, full-scale squarewave is shown in fig.7. The squarewave's shape is slightly different from that seen in other converters: the overshoot and ringing are lower in amplitude, and the ringing stops much sooner. Compare fig.7 to other converters' 1kHz, full-scale squarewave shapes. Although the No.30 uses the same digital filter as the Audio Research DAC1 and DAC1-20, their squarewave shapes are not identical, as might be expected.

Fig.7 Mark Levinson No.30, waveform of 1kHz squarewave at 0dBFS.

Output impedance was very low, measuring 6.8 ohms at 20Hz, rising to 7.9 ohms at 20kHz. At these very low output impedances, the cable connecting the device under test to the System One becomes a factor in the measurement. I suspect that the No.30 would meet the 6 ohm output impedance spec if measured directly at the output jacks, instead of connected to the System One. In any case, this is a very low value, indicating the No.30 will have little trouble driving any power amplifier, cable, or passive control unit. Balanced output impedance was similarly low, measuring 13.3 ohms at 20Hz, 13.6 ohms at 1kHz, and 15.4 ohms at 20kHz.

The No.30 doesn't invert absolute polarity unless the front-panel inversion switch is pressed. The balanced outputs assign pin 2 as "hot," conforming to the US standard. When connected to equipment in which pin 3 is "hot," the No.30 will invert absolute polarity when the front-panel inversion LED is off.

A minuscule level of DC was measured at its outputs: 100µV from the unbalanced jacks (both left and right channels), and 200µV (L) and 300µV (R) respectively from the balanced outputs. This is a very low level, especially considering that the No.30 is direct-coupled. It is not uncommon to measure several mV, or even tens of mV, of DC at the output of a digital processor. Incidentally, I couldn't get the front-panel AES/EBU LED to illuminate, even when driving it with a known AES/EBU signal from the Audio Precision System One Dual Domain.

Overall, the No.30's bench performance was exemplary. In many areas— crosstalk, linearity, de-emphasis error—the No.30 was the best-performing digital processor I've measured.—Robert Harley

Harman High Performance Audio/Video Group
1718 West Mishawaka Road
Elkhart, IN 46517
(516) 594-0300