Chord Electronics DAC64 D/A processor Measurements part 2
Fig.7 Chord DAC64, departure from linearity, 16-bit data (right channel dashed, 2dB/vertical div.).
Fig.8 Chord DAC64, waveform of undithered 1kHz sinewave at -90.31dBFS, 16-bit CD data.
The first sample of the DAC64 clipped above -1dBFS into the demanding 600 ohm test load. This was not true of the second sample, which could drive a 0dBFS signal into this load without raising a sweat (fig.9). Intermodulation levels were also low (fig.10).
Fig.9 Chord DAC64, spectrum of 50Hz sinewave, DC-1kHz, at -1dBFS into 600 ohms (linear frequency scale).
Fig.10 Chord DAC64, HF intermodulation spectrum, DC-25kHz, 19+20kHz at 0dBFS into 100k ohms (linear frequency scale).
The first sample's compromised dynamic range when fed 24-bit data was the first clue I had that something might be wrong with it. The second was when I used the Miller Audio Research Jitter Analyzer to examine its rejection of word-clock jitter. The grayed-out trace in fig.11 is a high-resolution spectral analysis of the DAC64's analog output while it decoded undithered 44.1kHz data representing a high-level tone at one quarter the sample rate, over which has been laid a low-frequency squarewave at the LSB level with the RAM buffer out of circuit. Not only is the analog noise floor a mess, with myriad spurious tones evident, spaced at 20Hz intervals, but the measured jitter level is a very high 6.32 nanoseconds! It's no wonder I disliked the first sample's sound quality without the RAM buffer engaged.
Fig.11 Chord DAC64, sample 1, high-resolution jitter spectrum of analog output signal, 44.1kHz sampling, PS Audio Lambda 2 transport via 6' Apature S/PDIF link (11.025kHz at -6dBFS with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Grayed-out trace is similar analysis without RAM buffer.
Switching in the buffer set to its small size reduced the jitter to a good 184 picoseconds, but setting the buffer to its maximum size increased the jitter, to 223ps, which was not what I was expecting. The increase was almost entirely due to a pair of sidebands at ±44Hz (each marked with a purple "2" in the foreground trace in fig.11), but the noise floor still looked dirty.
Fig.12 shows the results obtained under identical test conditions for the second sample of the DAC64. Again, the grayed-out trace is without the RAM buffer. There are still a number of spurious tones visible, but now the main characteristic is a large rise in the random noise floor on either side of the spike representing the 11.025kHz tone. The measured jitter level was a moderately high 587ps. Switching in the RAM buffer set to its maximum size (black trace) reduced the jitter to an excellent 169ps and eliminated the noise-floor peak. Most of the jitter is due to data-related sidebands (red numeric markers).
Fig.12 Chord DAC64, sample 2, high-resolution jitter spectrum of analog output signal, 44.1kHz sampling, PS Audio Lambda 2 transport via 6' Apature S/PDIF link (11.025kHz at -6dBFS with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Grayed-out trace is similar analysis without RAM buffer.
Finally, I am trying a different Miller Audio Research program that will allow me to measure word-clock jitter at sample rates other than 44.1kHz. Because it runs on a different National Instruments platform than the Jitter Analyzer, the results are not directly comparable with what I've been obtaining with the Analyzer. For interest's sake, however, fig.13 shows a similar spectral analysis taken with the new piece of kit while the DAC64 was decoding 96kHz Jitter Test data. The red trace was taken without any RAM buffer, the black trace with the maximum buffer. Again, the peculiar rise in the noise floor around the central tone can be seen without the buffer; again, the buffer drastically cleans this up, as well as reducing the level of high-frequency jitter components. However, it doesn't eliminate the lower-frequency components.
Fig.13 Chord DAC64, sample 2, high-resolution jitter spectrum of analog output signal, 96kHz sampling, PC fitted with RME Digi96/8 Pro soundcard via 1m TosLink (24kHz at -6dBFS with LSB toggled at 500Hz). Center frequency of trace, 24kHz; frequency range, ±3.5kHz. Red trace is similar analysis without RAM buffer.
Without a doubt, something was terribly wrong with the first sample of the Chord DAC64. Fortunately, the measured performance of the second sample was beyond reproach in almost every area. But I'm bothered by the apparent truncation of the 24-bit words when the RAM buffer is set to its maximum size. While this won't be an issue for playback of 16-bit CD data, I would have preferred not to have found it.—John Atkinson