Digital Processor Reviews

Sidebar 2: Measurements

The DA-10's unique ability to accept different DACs made measuring it more interesting and challenging. On one hand, it took nearly three times longer to measure. On the other, it was possible to look at the performance of different DACs with the same input receiver, filter, and output stage. Because presenting all the measurements on each DAC would consume too much space, I'll show the measurements with the AD1862 (the DAC most likely to be installed in the DA-10), and comment on the performance of the other DAC cards when of interest.

The DA-10's maximum output voltage when decoding a full-scale, 1kHz sinewave was 2.15V with the AD1862, 1.31V with the CS4328, and 3.39V with the UltraAnalog. Referenced to the standard 2V output, these levels are +0.63dB, –3.7dB, and +4.6dB respectively—a very large variance. When comparing the DAC cards, be sure to match levels with a voltmeter.

Output impedance—which is determined by the analog output stage, not the DAC—measured a low 109 ohms across the audio band. I measured very low levels of DC at the analog outputs: 300µV at the left channel, 400µV at the right. The DA-10 doesn't invert absolute polarity with any of the three DAC cards tested. It also had no trouble locking to the three standard sampling frequencies.

Frequency response (fig.1) was predictably flat, but the DA-10's de-emphasis error was unusual in that it changed when the I/V converter stage, which is used only with the AD1862 DAC card, was switched into the circuit. The middle pair of traces is the DA-10's de-emphasis error with the AD1862 card; the lower pair of traces is the de-emphasis error with the CS4328 card. With the UltraAnalog DAC, this error was identical to that measured with the CS4328, leading me to speculate that the I/V converter was somehow involved in this anomaly. I repeated these measurements several times on separate occasions, always obtaining identical results. The small positive error seen in the middle trace will cause a brighter, more forward presentation when playing pre-emphasized CDs. Although the error is small (less than half a dB), it spans more than three octaves of bandwidth—enough to be heard. The large negative de-emphasis error shown in the bottom traces will be much more audible: the rolloff is greater than 1dB at 10kHz, and 1.5dB at 20kHz. This will cause a slight loss of air and openness when playing pre-emphasized discs, perhaps accompanied by a reduction in immediacy. (Very few CDs are pre-emphasized, however (footnote 1).

Interchannel crosstalk was low, the DA-10's separation measuring 110dB at 1kHz, decreasing to 93dB at 20kHz. The crosstalk plot (fig.2) is dominated by power-supply noise at low frequencies. The peaks at 120Hz and 180Hz are power-supply noise intruding on the crosstalk measurement.

When I performed a spectral analysis on the DA-10's output when decoding a –90dB, 1kHz dithered sinewave, I saw a huge (15dB) linearity error in one channel. Trimming the MSB with the front-panel control reduced this error to 12dB, still a severe error. Because the AD1862 DAC card also has MSB trimmers next to the DACs (which are not intended to be user-adjustable), I trimmed the bad channel for no error at –90dB. This produced a spectral analysis that looked better, but didn't correct a fundamental problem with the DAC card. When I measured the DA-10's linearity, it indeed had no error at –90dB, but did have what looked like missing code transitions (seen in the linearity plot of fig.3). Notice that at –90dB (the level at which I set the internal MSB trimmer) there's no error, but clearly the right-channel DAC is severely misbehaving. No amount of MSB adjustment could correct this problem.

Counterpoint sent me a new AD1862 DAC card. Its linearity is shown in the bottom pair of traces in fig.4. The right channel's linearity is excellent, but the left channel has nearly 4dB of positive error at 90dB—hardly good performance. This linearity problem can't be corrected with the single front-panel MSB trimmer: if you set the left channel correctly, the right will be wrong. I then adjusted the left channel's internal MSB trimmer on the AD1862 DAC card so it matched the right channel, producing the top pair of traces in fig.4. Note that the top traces have been offset by +4dB, and the lower traces by –6dB, so I could show them on the same graph.

In short, the AD1862 DAC card, as shipped, didn't have its linearity matched between channels. Now we know that Counterpoint doesn't cherry-pick review samples. Two minutes on an Audio Precision System 1—JA tells me that Counterpoint does use a large number of Audio Precision systems for its production-line QA and setup—could have adjusted the internal MSB trimmers so that the DACs would have behaved identically. The failure of two samples in a row to perform optimally—samples sent for review and measurement—raises questions about Counterpoint's quality control which I assume they will address in their "Manufacturer's Comment."

Incidentally, don't try to adjust the internal MSB trimmers without proper test equipment; the range of the pot is so wide that severe misadjustment could easily occur. The front-panel MSB trimmer has a much smaller range; even if it's set incorrectly, the absolute linearity error won't be large.

Linearity measurements on the CS4328 and UltraAnalog DAC (not shown) indicated that these DACs' low-level performances were swamped by noise. The positive error at low levels suggested that the implementation of these DACs in the DA-10 was perhaps less ideal than that of the AD1862. Both the Crystal CS4328 and UltraAnalog D20400 have intrinsically better linearity and are quieter than I measured in the DA-10.

Fig.5 is the spectral analysis of a –90dB, dithered 1kHz sinewave with the AD1862 DAC card (after I set the internal MSB trimmer). The DACs are now well-behaved, although there's still a disturbing amount of power-supply noise intruding on the audio signal. This noise was there regardless of the DAC card installed, and remained even with different grounding arrangements.

The DA-10's reproduction of a –90dB, undithered 1kHz sinewave is shown in fig.6 from the AD1862 DAC, and in fig.7 from the UltraAnalog DAC. Note the much better wave shape and symmetry from the UltraAnalog DAC. The three transitions at this level (+1, 0, –1) are more clearly delineated in fig.7. Note how the power-supply noise seen earlier shifts the zero crossing point above and below the DC line. In effect, the 1kHz sinewave is riding on the low-frequency noise component.

Fig.8 shows the DA-10's excellent noise-modulation performance: the traces are tightly grouped, the noise level fairly low. This measurement was made on the best channel of the second sample, which I hand-trimmed with the Audio Precision.

The DA-10 intermodulation spectrum is shown in fig.9 with the AD1862, and in fig.10 with the UltraAnalog DAC. Although both spectra are quite clean and free from intermodulation components, the 1kHz difference component is completely nonexistent with the UltraAnalog DAC, but sticks up to –95dB with the AD1862. There are also IM components at multiples of 1kHz in the AD1862 FFT. The noise floor is also higher with the UltraAnalog DAC, even though it has a higher intrinsic signal/noise ratio than the AD1862. This further supports my assumption made earlier that this particular implementation of the UltraAnalog DAC degrades its noise performance, and that the DA-10's circuit works best with the AD1862.

Fig.11 is the 8x (352.8kHz) word-clock jitter when the DA-10 was driven by a full-scale, 1kHz sinewave. The signal-related jitter components are clearly apparent as spikes at odd multiples of 1kHz. The RMS level was a rather high 510 picoseconds, measured over a 400Hz–22kHz bandwidth. With a –90dB, 1kHz sinewave input, the jitter spectrum was much cleaner (fig.12), and the RMS level dropped to 450ps. Most processors have more, and higher-amplitude, signal-correlated jitter components at low signal levels. At –70dB (not shown), the spectrum looked very much like the full-scale spectrum shown in fig.11. To look at how well the DA-10 attenuates (or passes) higher-frequency jitter, I drove it with a full-scale, 10kHz sinewave. The spectrum, shown in fig.13, shows a large spike of jitter energy at 10kHz. Overall, the DA-10's jitter performance was less good than I have found to be possible with the CS8412 input receiver. (For comparison, see my measurements of the PS Audio Reference Link's jitter in Vol.16 No.10, p.206.)

Overall, the DA-10 was somewhat disappointing on the test bench. I've come to expect better technical performance from digital processors in this price range. Moreover, after receiving two misbehaving AD1862 DAC cards, I'm concerned about Counterpoint's quality control.—Robert Harley

---

Footnote 1: CD pre-emphasis is a high-frequency boost applied during recording or CD premastering. Pre-emphasized CDs carry a flag (in the Q-channel subcode) that tells the CD player or digital processor to switch in the de-emphasis circuit that attenuates the treble and restores flat response. If the de-emphasis circuit's attenuation characteristics don't match the pre-emphasis characteristics, de-emphasis error results. De-emphasis error can be thought of as a frequency-response error whenever the processor is decoding pre-emphasized discs.

Emphasis was designed to improve the CD's S/N ratio. By attenuating the treble after the DAC, DAC noise and artifacts are also attenuated. Very few discs have been pre-emphasized, particularly recently manufactured CDs, because engineers don't like the reduction in headroom it leads to. Although emphasis can subject the signal to more electronics (on both the recording and playback sides), it can theoretically improve CD performance if properly implemented.