Meitner IDAT D/A processor Measurements
Sidebar 3: Measurements
As John Atkinson wrote in the December 1992 issue of Stereophile about the relationship between measurement and sound quality ("If It Sounds Good..." p.15) a product should offer good engineering as well as good sound. This was certainly the case with the IDAT: its measured performance was superb.
The IDAT's output level when decoding a full-scale, 1kHz sinewave was 3.094V (left channel) and 3.122V (right) when driving a 100k ohm load. This is nearly 3.8dB higher than the standard 2V output level. From the balanced outputs, I measured 4.669V (left channel) and 4.664V (right). Note the virtually perfect channel balance. Output impedance was 387 ohms from the unbalanced outputs and 541 ohms from the balanced jacks at any frequency. DC levels were very low: unmeasurable at the left-channel unbalanced output, 2mV at the right. The balanced outputs had 1mV of DC at the right-channel jack and 2.1mV at the left.
The IDAT does not invert absolute polarity unless the front LED is illuminated. The unit had no problem locking to sampling frequencies other than 44.1kHz, but the 32kHz LED didn't illuminate when locked.
Unless noted, the following measurements were taken from the IDAT's balanced outputs.
Frequency response, shown in fig.1, revealed a 0.45dB rolloff at 20kHz, typical of most digital processors. The Wadia processors, whose DSP-based filters are optimized for the time domain, exhibit a rapid rolloff in the top octave (the Wadia 2000 is down 3dB at 20kHz). De-emphasis error was negligible (fig.1, lower trace), as would be expected; the IDAT performs de-emphasis in the digital domain with the DSP chips.
Channel separation (fig.2) was also excellent, measuring 125dB at 1kHz and 116dB at 20kHz. It's likely that this is the system's noise floor rather than crosstalk between channels; the IDAT's audio channels are housed in completely separate enclosures. Moreover, the virtually identical tracking between the LR and RL traces above 1kHz further suggests this hypothesis is correct.
A third-octave spectral analysis of the IDAT's output when decoding "digital silence" (fig.3) shows no audioband idle tones or other spuriae. The overall noise level is also low.
Performing a similar spectral analysis of the IDAT's output when decoding a 1kHz, 90dB dithered sinewave produced the plot in fig.4. There is a complete absence of power-supplyrelated noise (60Hz, 120Hz), a low overall noise level, perfect tracking between channels (revealed by the fact that the left-and right-channel traces closely overlap), and good linearity (the peak just reaches the 90dB horizontal division).
A better look at the IDAT's linearity is provided by fig.5. This is virtually perfect performance: the IDAT maintains its low-level linearity well below 100dB. There is a slight positive error that corrects itself, resulting in a 0.16dB and 0.65dB error in the left and right channels, respectively, at 112dB. This is among the best low-level linearity performances I've measured. Note that fig.5 was measured at the balanced outputs, which benefit from the fact that the whole DAC and analog stage is balanced, resulting in common-mode rejection of DAC non-linearity.
Interestingly, the IDAT's linearity measured at the single-ended outputs was quite different (fig.6). This is unusual in that the single-ended signal is derived from the balanced signal, and thus benefits from balanced DACs. Perhaps Ed Meitner can explain this in his "Manufacturer's Comments."
Another method of examining a DAC's low-level performance is to look at the converter's reproduction of very-low-level waveforms. Fig.7 is the IDAT's reproduction of a 1kHz, 90dB undithered sinewave. This is astonishing; I've never seen such a "textbook" waveshape. The quantization steps are perfectly shaped, uniform, and the waveform is overlaid with very little noise. This is the best-looking waveform I've seen, and very close to the theoretically perfect waveshape of a (nonexistent) ideal DAC shown in fig.8.
The IDAT's noise-modulation plot was also excellent (fig.9), with tight trace groupings. This reveals that the IDAT's noise floor, and the noise floor's spectral distribution, vary little with input level.
Performing an FFT on the IDAT's output when processing a full-scale mix of 19kHz and 20kHz produced the spectrum of fig.10. The test reveals the presence of intermodulation products generated by the device under test. The IDAT's 1kHz difference component (20kHz minus 19kHz) is noticeable at 90dB. There are also slight spikes at 3kHz, 6kHz, 11kHz, 13kHz, and each multiple of 1kHz above 13kHz. These are all low in level (below 100dB), but this is not exemplary performance on this test. The presence of these intermodulation products may correlate with my impressions of a slight hardness in the upper mids and treble.
The IDAT's reproduction of a 1kHz squarewave was unique: it doesn't suffer from time-domain distortions imposed by conventional digital filters. These distortions are seen as overshoot and ringing on the waveform. The IDAT's squarewave output, seen in fig.11, is virtually perfect, with no time-domain distortions. In fact, the squarewave looks as if it came directly from a signal generator, not a digital converter. It's difficult to overstate the achievement of designing a digital processor with such good time-domain performance.
As described in January's jitter article (pp.114145), we can now measure the level and spectrum of jitter in digital processors. As shown in fig.12, an FFT-derived spectrum analysis of the IDAT's jitter measured at the DAC word clock, the IDAT had extremely low jitter (78ps82ps RMS) and no discrete-frequency jitter components. It should also be noted that these jitter levels are very close to the LIM Detector's noise floor (when measuring an 8x-oversampling word clock). The IDAT's jitter figures may thus be lower than those presented here.
To assess the IDAT's C-Lock R (receiver) circuit on the digital input and the C-Lock T (transmit) on the digital outputs, I performed an experiment. I drove a Bitwise Musik System Zero processor with the code representing a 1kHz, full-scale squarewave from a JVC transport and coaxial connection, measured the jitter level, and took an FFT of the jitter spectrum. The result is shown in fig.13. Next, I inserted the IDAT in the digital signal path (with its C-Lock input and C-Lock output circuits) and performed the same tests on the Musik System Zero. The result was a huge reduction in jitter measured at the Musik System Zero's DAC from 3.46ns to 2.01ns (the 3.46ns jitter level is higher than that quoted in the January jitter article for the Zero; in those tests, I didn't measure jitter level when driven by squarewave data). In addition to significantly decreasing the overall RMS jitter level, the C-Lock circuits reduced the number and amplitude of discrete-frequency jitter components, shown in fig.14 (with C-Lock). Note the more random nature of the jitter and the lower overall level. C-Lock appears to work as claimed.
In short, the IDAT's measured performance was superb. The processor's low-level performance was particularly good, shown by the linearity plot and the "textbook" waveshape of a 1kHz, 90dB undithered sinewave. The IDAT's virtually perfect squarewave reproduction is a remarkable achievement. Finally, the IDAT's low jitter level and lack of periodic jitter and LIM components are exemplary.Robert Harley