Digital Audio Labs CardDeluxe PC soundcard Measurements
I looked first at the DAL CardDeluxe's A/D converters. The analog input impedance was 10.8k ohms across the audioband, dropping very slightly at 20kHz. With the card's jumpers set to the higher of the two output levels (this is how I did all my auditioning), the analog input clipped at 7V RMS in, equivalent to 7.06V out, and 16.9dB above a nominal 1V output level. Fig.1 shows the complete A/D/A frequency response at three sample rates: 32kHz, 48kHz, and 96kHz. The audioband response is flat, but at the highest sample rate the top-octave output droops a little at both frequency extremes.
Fig.1 Digital Audio Labs CardDeluxe, A/D/A frequency response at -1dBFS with 32, 48, and 96kHz sampling (right channel dashed, 2dB/vertical div.).
The CardDeluxe converters output 24-bit data words, but without cryogenic freezing of the circuits and ultra-low impedance topologies, there isn't an analog circuit yet designed that can achieve a full 24 bits of dynamic range. Not that this matters—while the practical limit that can be currently achieved is around 20 bits, digital data with a true 20 bits of resolution should be sonically transparent to all listeners under all conditions.
The red trace in fig.2 shows a high-resolution spectrum of the digital data produced by the CardDeluxe's A/D converter driven by a 1kHz analog sinewave from my Audio Precision System One. (The digital data were analyzed by a PrismSound DScope II, which has sufficient mathematical dynamic range to examine 24-bit data.) The noise is below -140dBFS, and though distortion components can be seen at two, four, five, six, seven, and eleven times the fundamental frequency, the highest in level of these—the second harmonic at 2kHz—is still an astonishing 111dB below full-scale. However, sidebands at ±250Hz can be seen surrounding the 1kHz peak, as well as some lower-frequency components, including one at 60Hz. But at -113dB, this is of only academic interest.
Fig.2 Digital Audio Labs CardDeluxe, A/D spectrum of 1kHz sinewave, DC-22kHz, at -1.7dBFS (red trace); and of digitally synthesized, TPDF-dithered, 20-bit 1kHz sinewave (blue). (Linear frequency scale, 10dB/vertical div.)
Though fig.2 indicates better than 16-bit dynamic range for the CardDeluxe's analog inputs, it is not quite up to the best. The blue trace in fig.2 shows a spectral analysis of a 1kHz tone generated in the digital domain by the DScope and dithered to 20-bit word length. Its noise floor lies approximately 11dB below that of the CardDeluxe's A/D converter, suggesting that the latter achieves slightly better than 18 bits' dynamic range. This is still excellent, and some of the analog noise might indeed be due to the Audio Precision generator.
The digital input appeared to be transparent to the input data; ie, the data words stored on the computer's hard drive were the same as output by the digital source. (I understand that this is not always the case with inexpensive soundcards.)
Turning to the balanced analog outputs, the source impedance was 598 ohms across the audioband, and absolute polarity was correct with XLR inputs wired with pin 2 hot. Channel separation was better than 120dB below 1kHz, and though some slight capacitive coupling between channels reduced this to around 100dB at 20kHz, this is still superb performance.
To test the D/A converters, I mainly used 44.1kHz WAV files played back with CoolEdit 2000. This gave one anomaly, visible in fig.3—a graph of the CardDeluxe's frequency response with and without pre-emphasized data. While the response is superbly flat with regular data, driving the card with pre-emphasized data, both from the WAV file and from an external CD transport, gave the top response, which shows that the appropriate de-emphasis was not being applied. I checked that the "emphasis" flag was correctly set in the external source datastream. It was, suggesting that this flag was not being recognized by either the CardDeluxe or CoolEdit 2000. But given that the chances of the CardDeluxe being used to decode pre-emphasized data are very small, I don't see this as a serious problem.
Fig.3 Digital Audio Labs CardDeluxe, D/A frequency response with (top) and without (bottom) de-emphasis at -12dBFS, 44.1kHz sampling (right channel dashed, 2dB/vertical div.).