Quality Lies in the Details Page 4

Frequency response
Because the errors are much smaller, frequency response doesn't usually reveal clues about a processor's overall tonal balance, as it often does with loudspeakers. We include it anyway, in case anything unusual is going on in the unit under test. Most processors are down a few tenths of a decibel at 20kHz.

More revealing, however, is a CD player's or processor's de-emphasis error. Some CDs have had the treble boosted (emphasized) during recording, and carry a flag in the datastream that tells the CD player to switch-in its de-emphasis circuit to restore flat response. These de-emphasis circuits are often inaccurate (due to resistor and capacitor tolerances), producing frequency-response irregularities when the player is decoding emphasized discs. De-emphasis error should be less than about 0.1dB across the band.

We measure de-emphasis error by driving the processor with a flat, swept sinewave (100Hz-20kHz) from the digital signal generator, and setting the flag in the datastream to "Emphasis On." This causes the processor to engage its de-emphasis circuit and roll off the treble. The measured rolloff is compared to the exact rolloff specified by the CD Standard, and the difference between them is plotted. Then, the processor's frequency-response errors are subtracted to show only the de-emphasis error—not the de-emphasis error superimposed on the frequency-response errors. The resulting curve is thus only the de-emphasis error, and should ideally be a flat line (footnote 3). De-emphasis errors can cause tonal-balance irregularities (tizzy upper treble, for example) when playing pre-emphasized CDs. De-emphasis errors of 0.2dB over an octave of bandwidth are audible.

Channel separation
The separation between a player's stereo channels—often referred to by its reciprocal, the interchannel crosstalk—is a measurement of how well the left and right channels are isolated from one another. Ideally, when a digital processor is fed a right-channel signal, that signal should not appear in the left channel, and vice versa. In practice, a small amount, called the crosstalk, does leak from one channel to the other. The lower the crosstalk—or, put another way, the greater the channel separation—the better. Many processors (and other products) have decreasing channel separation as frequency increases—a result of the fact that adjacent wires or printed-circuit-board traces act as a capacitor between the channels.

Here's how we measure channel separation: The processor is driven in one channel by a full-scale signal that sweeps from 100Hz to 20kHz. The Audio Precision System One measures the signal present in the undriven channel (the crosstalk), and plots that signal as a function of frequency. This is repeated, with the driven and measured channels reversed. Typical CD players and digital processors have about 90dB of channel separation at 1kHz, decreasing to about 70dB at 20kHz.

Although there is no correlation between low crosstalk and soundstage width (an intuitive link), high channel separation indicates good overall engineering (footnote 4). At extremely low levels (the 135dB separation measured in the Mark Levinson No.30, for example), it gets to the point where the device we're trying to measure is actually measuring the Audio Precision! Typical excellent and average crosstalk measurements are shown in fig.4.

Fig.4 Excellent (bottom) and average (top) digital-processor crosstalk performance (L-R dashed, 10dB/vertical div.).

Spectral analysis
The next measurement we routinely perform is a spectral analysis of the processor's output when the processor under test is driven with the digital code representing a dithered 1kHz sinewave at -90dBFS. (The word "dithered," explained in past issues of the magazine, means that a very small amount of noise has been used at the encoding stage to ensure that this sinewave is free from distortion.) A bandpass filter 1/3-octave wide is swept over the audioband. The resultant spectral analysis graph plots level vs frequency. We look for peaks at 60Hz, 120Hz, 180Hz, and related frequencies that indicate whether power-supply noise is getting into the audio circuitry. One CD player I recently measured (fig.5) had lots of power-supply noise in one channel, but almost none in the other. The channel with the noise was located right next to the power transformer; the noiseless channel was on the other side of the chassis.

Fig.5 Typical spectrum of dithered 1kHz tone at -90.31dBFS, with noise and spuriae (1/3-octave analysis, right channel dashed). Note that the left-channel spectrum has power-supply components visible at 60Hz and its harmonics.

This spectral analysis also hints at how well the unit's DACs (digital/analog converters) are performing: the signal should peak at the -90dB horizontal line, and the left and right traces should closely overlap. This indicates that the DACs are linear and have similar performance between channels.

Footnote 3: This isn't as difficult as it sounds. I wrote a procedure for the System One that performs all these functions automatically and saves the test data on a floppy disk. In fact, nearly all the digital processor tests described are executed automatically (and identically each time) by the computer according to custom-written instructions.

Footnote 4: A phono cartridge has (optimistically) perhaps 35dB of channel separation at 1kHz, yet there is no constriction of soundstage width from LPs.

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?