MSB Technology Link D/A converter Measurements part 2
Fig.4 shows the waveform of a 1kHz squarewave sampled at 96kHz. As with the California Audio Labs CL-20, the ringing on the waveform tops is higher in frequency than with CD data. Note the slight leading-edge droop in this graph, which correlates with the small amount of ultrasonic rolloff apparent in fig.3.
Fig.4 MSB Link, 1kHz squarewave at 0dBFS, 96kHz sampling.
So, the 96kHz part of the MSB's specification operates. How about the 24-bit part? Fig.5 shows the spectrum of the Link's output while it decodes data representing a dithered 1kHz sinewave at -90dBFS. The top traces are with the generator's word length set to 16 bits, the bottom traces with it set to 24 bits. There is an approximately 8dB reduction in the noise floor, suggesting a dynamic range equivalent to between 17 and 18 bits. Note that the MSB's resolution—defined as how its output level changes when one or more LSBs changes value—is probably better than this. In any case, this is astonishingly good performance for a processor costing just $350!
Fig.5 MSB Link, spectrum of dithered 1kHz tone at -90.31dBFS, with noise and spuriae: 16-bit data top, 24-bit data bottom (right channel dashed).
Extending the measurement bandwidth to 200kHz and using digital silence as the test signal gave the spectra shown in fig.6. Again, increasing the word length, this time to 20 bits, dropped the audioband noise floor by 8dB or so. But note the huge increase in noise above 30kHz. This is due to the massive noise-shaping Burr-Brown has employed to wrest maximum resolution out of the Link's DAC. Some preamps and power amps might misbehave when fed this kind of ultrasonic spectrum.
Fig.6 MSB Link, spectrum of digital silence with noise and spuriae: 16-bit data top, 20-bit data bottom (1/3-octave analysis, right channel dashed).