Digital Processor Reviews

Before examining the DEQX PreMate's equalization and correction features, I wanted to make sure that its basic performance as a D/A processor was up to the expected standard—"First, do no harm" is as fundamentally important in digital signal processing as it is in healthcare. I examined the PreMate's electrical performance with my Audio Precision SYS2722 system (see www.ap.com and the January 2008 "As We See It"), mainly using the TosLink optical input, though I repeated some of the tests using a USB connection from my MacBook Pro laptop running on battery power. Apple's USB Prober utility identified the processor as the "XMOS USB Audio 2.0" and confirmed that the USB input operated in the optimal isochronous asynchronous mode. The USB input accepted 24-bit data with sample rates ranging from 44.1 to 192kHz, including 88.2 and 176.4kHz. The TosLink input accepted data with sample rates up to 96kHz, and the AES/EBU input sample rates up to 192kHz. However, the DEQX Calibration app (v.2.93), running on a Windows XP machine, indicated that 192kHz data were downsampled to 96kHz, and 176.4 down to 88.2.

Before performing any measurements, I selected the P0 preset with the remote control, which, as programmed by KR, sets all DSP functions to flat, though the DSP engine remains in the data path. I also pressed EQ Reset on the remote, to ensure that the parametric equalizer was set to do nothing. With the digital inputs' sensitivity set to "0dB" with the DEQX Cal software, the maximum output level at 1kHz was 2.847V from the balanced outputs, 1.327V from the unbalanced outputs. (When shipped, the PreMate is set-up with 6dB of digital headroom when the analog output gain is set to 0dB.) A 0dBFS signal was indicated as "+0.1dB" on the DEQX Cal's meters. With the signal polarity set to "+" on the software's control panel, the PreMate inverted signal polarity from both sets of outputs. The balanced output impedance was 200 ohms at high and middle frequencies, rising to 1260 ohms at 20Hz. The unbalanced output impedance was 74 ohms across the audioband.

The impulse response of the PreMate's digital reconstruction filter is shown in fig.1. Unusually but commendably, it appears to have a minimum-phase characteristic, with no pre-ringing. Fig.2 shows the results of the test suggested to me by Jürgen Reis of MBL, in which the device under test is fed 44.1kHz-sampled data first representing a full-scale 19.1kHz tone, then white noise at –4dBFS. The reconstruction filter rolls off sharply above 21kHz, and while it hasn't reached full attenuation at the Nyquist frequency (half the sample rate, indicated by the vertical green line), the image of the tone at 25kHz is completely suppressed. The distortion harmonics of the tone are very low in level, but there is a peculiar rise in the noise floor to either side of the spectral spike that represents the tone, as well as symmetrical low-level sidebands, both presumably due to jitter of some kind.

Fig.3 shows the conventional frequency response of the PreMate's digital inputs, plotted at 44.1 and 96kHz with the P0 preset. (I haven't shown the 192kHz response, as it was identical to that at 96kHz, which suggests that the DSP signal path operates at a maximum rate of 96kHz.) There is a smooth ultrasonic rolloff, followed by a sharp drop just before each Nyquist frequency. However, the PreMate will not generally be used in its bypass mode, but instead will be programmed to apply equalization. Fig.4 shows one of the programmed filter responses Kal Rubinson had used with the Bryston Middle T loudspeakers; fig.5 shows the measured response. The theoretical and actual responses are, as hoped, identical, other than the reconstruction filter's slow rolloff above the audioband.

Channel separation was excellent, at >115dB in both directions below 1kHz, and still 95dB at 20kHz. The noise floor (fig.6) was free from power-supply–related artifacts. To investigate the PreMate's ultimate resolution, I used the DEQX Cal app to remove the 6dB of headroom. I then fed it dithered data representing a 1kHz tone at –90dBFS with first a 16-bit word length, then 24 bits, using the TosLink input. The results are shown in fig.7. The increase in bit depth lowers the noise floor by 18dB, implying that the PreMate offers 19-bit resolution, which is excellent. Repeating the test with the USB input gave identical results, confirming that the USB receiver correctly handles 24-bit data.

When a digital signal-processing engine is used to apply gain, as the signal data can't exceed 0dBFS, it actually attenuates the signal at frequencies at which gain is not required. This will reduce the ultimate resolution by the amount of attenuation applied. Fig.8 repeats the resolution analysis using KR's filter setting examined in figs.4 and 5. The DEQX app revealed that the gain offset was now 3.1dB in the left channel, 2.5dB, right, and the 24-bit noise floor had risen by up to 8dB. Resolution was thus approximately one bit lower than when the PreMate was set to do nothing, but the noise floor was commendably free from idle tones and other spuriae. The PreMate's reproduction of an undithered tone at exactly –90.31dBFS clearly shows the three DC voltage levels described by the data (fig.9), with the waveform symmetry implying very low DAC linearity error. With undithered 24-bit data, the result was a well-defined sinewave (fig.10).

Harmonic distortion was very low, even into the demanding 600 ohm load (fig.11), with the second harmonic the highest in level at –97dB (0.0014%) and –106dB (0.0005%) for the right and left channels, respectively. Intermodulation distortion was also very low, but the peculiar rise in the noise floor to either side of the high-level tones evident in fig.2 is also visible in fig.12. It was only in its rejection of jitter that the PreMate fell short of excellent performance. Fed 16-bit J-Test data via TosLink or USB, the result was the same (fig.13): the noise-floor modulation, accompanied by many low-level sidebands that obscure the high harmonics of the LSB-level, low-frequency squarewave. (The green line shows the expected level of these harmonics.) The picture didn't change with 24-bit J-Test data.

Other than the peculiar behavior of its noise floor with high-level, high-frequency signals, the DEQX PreMate, considered as a D/A processor, offers superb measured performance. That it also has an enormously flexible DSP engine is a bonus!—John Atkinson