Quality Lies in the Details Page 6

Low-level resolution
Related to linearity performance is how well a DAC can reproduce a very low-level signal. One way to examine a DAC's behavior on small signals is to capture its analog output waveform when reproducing a -90dB, undithered 1kHz sinewave. This test signal produces three quantization steps: 0, +1, and -1. The three levels should be of equal amplitude, and the signal should be symmetrical around the center horizontal division. Fig.8 shows the ideal waveshape (assuming an infinite bandwidth); a nearly perfect reproduction of this signal is shown in fig.9, which overlays the fig.8 waveform with small amounts of Gibbs Phenomenon ringing at the bit transitions owing to the CD system's limited, 22kHz bandwidth. (The higher harmonics of the signal, which would otherwise give sharp waveform "corners," are suppressed.) Tilts in the trace reveal the presence of low-frequency noise and power-supply problems—see my Jadis measurements elsewhere in this issue—while high levels of audioband noise result in poor reproduction of the true waveform (fig.10).

Fig.8 Perfect waveform of undithered, 16-bit, 1kHz sinewave at -90.31dBFS with no bandwidth restriction.

Fig.9 Excellent CD player, waveform of undithered, 16-bit, 1kHz sinewave at -90.31dBFS. The "ringing" at the bit transitions is almost entirely due to the CD system's restricted bandwidth.

Fig.10 Poor CD player, waveform of undithered, 16-bit, 1kHz sinewave at -90.31dBFS. The bit transitions are smothered by audioband noise.

Noise modulation
Testing a DAC's noise modulation was suggested by an Audio Engineering Society paper given by Dr. Richard Cabot, Principal Engineer at Audio Precision (footnote 5). The measurement looks at how a digital processor's noise floor shifts (modulates) with changing input level. Not only does the noise floor change in level, but the noise floor's spectral balance can be very different at different input levels. Research by Louis Fielder at Dolby Laboratories (those guys know something about noise modulation!) suggests that spectral shifts as small as 2dB are audible.

In the noise-modulation test, the processor is driven by the code representing a 41Hz sinewave at -60dBFS. The low-frequency test signal exercises the DAC, and the resulting noise after the processor's output has been high-pass-filtered to remove the test signal is plotted against frequency. The test is repeated at four additional levels (-70dB, -80dB, -90dB, -100dB), and the plots are combined on a single graph. This makes comparing noise-floor levels and spectral balance between input levels easier.

Footnote 5: "Noise Modulation in Digital Audio Devices," presented at the 90th AES Convention, Paris, 1991.

hollowman's picture

Can one correlate intersample overs to the square wave phenomenon JA is referring to here? That is, Philips, Wadia, Pioneer's Legato Linear, and Meitner do not "compress" the overs?