The Theorem had the lowest output voltage of the processors I review this month, measuring 1.97V when reproducing a full-scale, 1kHz sinewave. Output impedance was a fairly low 152 ohms across the band, which suggests the Theorem will have no trouble driving a passive level control.
The unit's frequency response (fig.1) showed a moderate (0.3dB) rolloff at 20kHz. De-emphasis tracking (also shown in fig.1) was virtually perfect. Interchannel crosstalk, shown in fig.2, was difficult to measure; the true crosstalk didn't emerge from the noise floor until 5kHz and above, indicated by the flat, then gently rising curve.
A spectral analysis of the Theorem's output when decoding digital silence is shown in fig.3. There is a a fairly high level of 60Hz power-supply noise and an unusual peak at 50kHz, either the result of the noise-shaping scheme in the Burr-Brown PCM67 hybrid DAC or due to an idle tone. Fig.4 is the same type of spectral analysis, but this time with the Theorem decoding a dithered 1kHz, –90dB sinewave. There is a small peak at 2kHz in the left channel not present when the Theorem decoded digital silence, suggesting it is a signal-related second-harmonic distortion. It is, however, very low in level.
Low-level linearity (fig.5) was only fair, with a 2.5dB positive error (right channel) and a 1.7dB positive error (left) at –90dB. The channels are not that well matched, especially considering that both converters are within the same monolithic chip.
The noise-modulation plot (fig.6) reflects the Theorem's less than ideal low-level linearity. There is a fairly wide divergence in the traces, particularly above 3kHz. Moreover, the bandwidth of the change in the noise floor's spectral balance as a function of input level is wide, spanning two and a half octaves. Capturing the Theorem's reproduction of a 1kHz, –90dB undithered sinewave produced the waveform of fig.7. There seems to be a low level of audioband noise, and inexplicably, the fourth and fifth positive peaks in the plot are of higher amplitude. The low noise level is paradoxical: the Theorem had a subjectively higher noise level than the other converters, and the crosstalk measurement also seemed to indicate a high noise floor. The unusual shape of this waveform is perhaps idiosyncratic to the hybrid DAC used in the Theorem; this is the first processor I've measured to use this hybrid converter.
An FFT-derived spectral analysis of the Theorem's output when decoding a full-scale mix of 19kHz and 20kHz is shown in fig.8. The 1kHz difference component and sidebands around the test signals are well down in level, but there are a few low-amplitude spikes in the audioband. These spikes, however, are far fewer and lower in level than those seen in the Fort;ae DAC 50's intermodulation spectrum. Fig.9 is the Theorem's reproduction of a 1kHz, full-scale squarewave. It has slightly less overshoot and more ringing than the Fort;ae DAC 50, but is otherwise typical in shape.
Finally, the Theorem doesn't invert absolute polarity, and I measured 200µV of DC offset (left channel) but a moderate 1.2mV offset at the right-channel output.
Compared with the other processors, the Theorem's bench performance was only mediocre. There was, however, nothing in the plots that would indicate the Theorem's musicality.—Robert Harley
