Chord Electronics DAC64 D/A processor Measurements
Unless indicated otherwise, the test results refer to the second sample. The maximum output levels were a high 3.097V RMS from the unbalanced jacks (this 3.75dB higher than the CD standard's 2V) and 6.197V from the balanced. The unit inverted absolute polarity from the XLRs with pin 2 wired hot; it was non-inverting from the RCAs. Source impedance was a very low 67 ohms (unbalanced) and 135 ohms (balanced) across most of the audioband, these figures rising at 20kHz to a still low 99 ohms and 207 ohms, respectively.
The frequency response with CD data was flat from both sets of outputs (fig.1, top traces), and was not changed by the RAM buffer. The response error playing back pre-emphasized data was minimal (fig.1, bottom traces). Driving the DAC64 at 32kHz, 48kHz, and 96kHz sample rates with the RAM buffer engaged gave the responses in fig.2, all perfectly straightforward (other than the fact that I had to compensate for the buffer's time delay when I did the test). Channel separation (not shown) was superb, at better than 110dB in the midrange, though this did decrease to a still excellent 95dB at 20kHz due to capacitive coupling.
Fig.1 Chord DAC64, frequency response at -12dBFS, 44.1kHz sampling, without emphasis (top) and with emphasis (bottom). (Right channel dashed, 0.5dB/vertical div.)
Fig.2 Chord DAC64, frequency response at -12dBFS with (from left to right): 32kHz, 48kHz, and 96kHz sampling (right channel dashed, 1dB/vertical div.).
Fig.3 shows spectral analyses of the first sample's noise floor with the RAM buffer in-circuit, driven with 16- and 24-bit dithered data representing a 1kHz tone at -90dBFS. The traces are free from spuriae, but the 8-bit increase in word length lowers the noise floor by only 6dB, or one bit's worth of dynamic range—poor performance by modern DAC standards. Repeating the spectral analyses on the second sample gave the traces shown in fig.4. Now the increase in word length drops the noise floor by 10dB across the band, implying almost 18-bit performance. While this is about one bit worse in the treble than the Musical Fidelity A324, the Chord maintains the improvement to the lowest frequencies, whereas the A324's noise floor is less good in the lower midrange and bass (fig.3 in that review).
Fig.3 Chord DAC64, sample 1, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS, with noise and spuriae (from top to bottom): 16-bit data, 24-bit data (right channel dashed).
Fig.4 Chord DAC64, sample 2, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS, with noise and spuriae (from top to bottom): 16-bit data, 24-bit data (right channel dashed).
The RAM buffer was switched off for the measurements represented by figs.3 and 4. I was surprised to find that when I switched it in, it changed the result when it was set to its maximum size. The top two pairs of traces in fig.5 show spectral analyses of the DAC64's noise floor while it decoded a 24-bit dithered 1kHz tone at -90dBFS without the RAM buffer (same trace as the bottom trace in fig.4, if you allow for the different vertical scaling) and with the buffer set to its smaller setting: they overlay exactly. However, when the buffer was set to its maximum size, I got the bottom pair of traces in fig.5. The noise floor has dropped by around 10dB, and now there are some spurious harmonic tones apparent. It looks as if the dither noise has been removed, which would happen only if the RAM buffer was truncating the 24-bit data words by a couple of least-significant bits. A puzzle.
Fig.5 Chord DAC64, sample 2, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS, with noise and spuriae (from top to bottom): 24-bit data, no buffer; 24-bit data, small buffer; 24-bit data, large buffer (right channel dashed).
Fig.6 shows similar spectral plots but with the measurement bandwidth extended to 200kHz while the Chord decoded 16- and 24-bit data representing "digital black." The same across-the-audioband lowering of the noise floor by 10dB can be clearly seen.
Fig.6 Chord DAC64, 1/3-octave spectrum of "digital black," with noise and spuriae (from top to bottom): 16-bit data, 24-bit data (right channel dashed).