Sony DTC-2000ES DAT recorder Measurements
Because the DTC-2000 is the first consumer product with Super Bit Mapping, I was eager to look at its technical performance and investigate Sony's published claims and technical graphs of what SBM does. But first, let's look at the DTC-2000's D/A and A/D converter performance.
The unit's maximum output level was 2.5V when fed 0dBFS, 1kHz sinewave data. Output impedance was a moderate 295 ohms at any audio frequency. The DC levels at the output were very low, measuring just 0.9mV (left channel) and 0.1mV (right).
The line input impedance measured 43.6k ohms at 1kHzslightly below the 50k nominal value. The line input required a voltage of 1.3V RMS to produce a digital signal of 0dBFS with the input level control exactly halfway up, suggesting that the DTC-2000 will interface well with a variety of source levels. The DTC-1000 doesn't invert absolute polarity from analog input to output, or in its D/A section. I also found the DTC-2000's digital meters to be very accurate, precisely tracking those of the Audio Precision System One.
Fig.1 shows the DTC-2000's D/A frequency response (top traces), de-emphasis error (middle traces), and overall record/replay response from analog input to analog output (bottom traces). Although the responses are flat, the DTC-2000 has a mild de-emphasis error that will be audible as a slight darkening of the sound with pre-emphasized tapes. Although the 0.7dB maximum rolloff isn't severe, it covers a broad enough band to be just audible. Channel separation was excellent, measuring 106dB at 1kHz, rising gently to 90dB at 20kHz. The DTC-2000 produced low levels of intermodulation products when driven by a code representing an equal-level mix of 19kHz and 20kHz tones. The 1kHz difference product lay below 100dBexcellent performance.
Fig.2 is a spectral analysis of the DTC-2000's output when decoding a 90dB, 1kHz undithered sinewave. The spectrum is free from power-supply noise (except a trace of 60Hz in the right channel), and the overall noise level is low. (Compare this plot with that of the Bel Canto Aida, also reviewed in this issue.) Fig.3 is the DTC-2000's noise-modulation plot. This is superb performance, and among the best I've measured. The noise level is low, and the traces are very tightly grouped, almost appearing as a single trace above about 3kHz.
The DTC-2000's D/A converter linearity was excellent (bottom traces in fig.4), as would be expected from Sony's 1-bit type DAC. The A/D converter linearity (top traces in fig.4) was also good, but the apparent positive "linearity error" below 90dB is more likely noise swamping the signal. Nonetheless, these are both excellent linearity plots.
We can see this difference in noise performance between the D/A and A/D stages in figs.5 and 6. Fig.5 is the DTC-2000's reproduction of a 90dB, 1kHz undithered sinewave with a digital input driving the DTC-2000. This excellent waveshape is overlaid with very little audioband noise, and the transitions are uniform. For comparison, fig.6 shows a low-level waveform as reproduced by the DTC-2000 with a very-low-level analog input signal (50µV). We can see that the A/D converter adds a fair amount of noise to the signal, as would be expected; A/D converter technology is way behind D/A converter engineering.
Fig.6 was made with the Super Bit Mapping turned off. Fig.7 is the same test signal, but with SBM turned on. The much higher noise level seen with SBM is a result of the noise floor's spectral energy being shifted higher in frequency. The noise energy is so frequency-specific that you can almost count the cycles of the noise. (I count between 18 and 20 cycles overlaying each 1kHz wave, which suggests that the noise energy is concentrated at the upper end of the audio bandexactly what's expected from knowing what SBM does.)
There's been some debate over potential audible problems when so much noise energy is concentrated over a narrow band of frequencies. When noise energy is concentrated over too narrow a frequency band, it begins to be heard as a specific pitch. At the 1990 AES convention in Montreux, for example, JA attended a demonstration of an early noise-shaping technique. He immediately noticed a high-frequency "whistle" overlaying the music, the shaped noise acquiring too much character. [To be fair, I was listening to very-low-level music considerably louder than the noise-shaper algorithm's designers ever anticipated. It could be argued that, at lower playback levels, the shaped noise would have dropped below my hearing's high-frequency threshold and thus have lost its audible pitch.Ed.]
Looking again at fig.7, we can see the potential to hear that signal as a tone overlaid with a second tone; the signal almost looks like a twin tone of 1kHz and 20kHz. Note, however, that the ear's sensitivity drops at 20kHz, making the high-frequency noise component less audible than is suggested in this graph.
We can look more closely at the noise floor's shape by performing a 1/3-octave spectral analysis of the DTC-2000's output with no input signal. I did this with and without SBM engaged, producing the plots in fig.8. The curve made with SBM (solid trace) shows a lower noise floor throughout most of the audioband, but vastly higher noise in the top octave. We saw this high-frequency noise overlaying the waveform in fig.7. Note that noise-shaping techniques such as SBM can't lower the overall noise level within the audioband. Instead, they merely shift it around, as seen in this graph.
Sony's promotional graphs for SBM show apparently improved resolution with SBM on low-level waveformssee Stereophile, Vol.15 No.8, p.55. This is somewhat misleading, however. Sony's engineers invoked a 16kHz low-pass filter for their measurements, removing the concentration of high-frequency noise from the waveform. It could be argued, however, that human hearing acts as a low-pass filter (reduced sensitivity at very high audio frequencies), making the filtered waveforms subjectively appropriate.
At any rate, what matters is how SBM affects our musical impressions. And JGH did indeed think highly of its effect.Robert Harley