Sutherland Engineering 12dAX7 USB DAC/preamplifier Measurements

Sidebar 3: Measurements

To test the Sutherland 12dAX7, I used a Dell 866MHz Pentium 3 desktop PC fitted with both USB 1.0 and USB 2.0 ports, with the test signals in WAV format and output using CoolEdit 2000. Installing the device was simple. When the special symmetrical USB cable—with a flat connector at each end—was hooked up, the PC's new hardware wizard recognized the 12dAX7's Burr-Brown PCM2702 chip and installed the necessary driver files from the CD-R supplied by Sutherland. I could then choose the 12dAX7 as the output device from within CoolEdit 2000. I later used my Apple iBook to drive the Sutherland. An amber LED visible through the unit's transparent front panel glows green when USB data are present at the unit's input.

The maximum output level was 2.86V RMS, although this was available only into high impedances (see later). The 12dAX7 inverted polarity, and its output impedance was a moderate 872 ohms at 1kHz and 20kHz. However, at low frequencies the impedance rose significantly, reaching 7.3k ohms at 20Hz. The amplifier or powered speakers used with the Sutherland need to have an input impedance of at least 50k ohms if dynamics and bass extension are not to be compromised.

The 12dAX7's frequency response is shown in fig.1. It's basically flat across most of the audioband, though a rise in output can be seen above 4kHz, reaching a maximum boost of 1dB at 20kHz. This will be audible in side-by-side comparisons with other processors. I couldn't get the unit to recognize pre-emphasized data, so the response in this case showed the usual rise at high frequencies (not shown). However, it is extremely unlikely that a user will ever feed the Sutherland a pre-emphasized music file. Because the Burr-Brown chip used by the 12dAX7 conforms to the USB 1.0 specification, which has a maximum data rate of 12Mbps, it cannot handle audio data with a sample rate greater than 48kHz. What was odd, however, was that when I fed it 96kHz-sampled data, I did get an audio signal out, at the correct frequency. Perhaps the CoolEdit program downsamples on the fly when outputting USB data.

Fig.1 Sutherland 12dAX7, frequency response at -12dBFS into 100k ohms with volume control at maximum (right channel dashed, 0.5dB/vertical div.).

Fig.2 shows the channel separation plotted against frequency. Though very respectable in the midband, it decreases steadily with increasing frequency due to capacitive coupling—probably at the analog volume control, given the physical separation of the tubed output stages. (Although the PCM2702 has an on-chip digital volume control, it looks as if Ron Sutherland has used an analog-domain control.) However, the apparent decrease at low frequencies is not due to crosstalk per se, but to some 60Hz and 180Hz hum that I could not get rid of, no matter how I arranged the grounding of the computer, the Sutherland, or the Audio Precision System One test gear. I could get rid of it if I drove the Sutherland from my iBook on battery power, and I note that Mikey Fremer suffered from similar hum only when he first set up the unit, and not on subsequent occasions. It's possible, therefore, that the 12dAX7's hum pickup is very dependent on the computer with which it is used.

Fig.2 Sutherland 12dAX7, channel separation vs frequency (10dB/vertical div.).

The hum components can be seen in fig.3, which shows a spectral analysis of the Sutherland's analog output while it decodes 16-bit data representing a dithered 1kHz tone at -90dBFS. Note that the noise floor almost obscures the 1kHz peak, with the right channel being worse than the left, and that the left channel shows a tad of second-harmonic distortion. As the Burr-Brown PCM1702 is a 16-bit device, it will truncate digital audio data with bit depths greater than 16. Increasing the word length to 24 bits and repeating the fig.3 measurement gave no change in behavior, therefore (not shown).

Fig.3 Sutherland 12dAX7, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS, with noise and spuriae (16-bit data, right channel dashed).

Because of the relatively high levels of analog noise and hum, the linearity error became increasingly positive at levels below -70dBFS (fig.4); the waveform of an undithered 1kHz tone at -90.31dBFS was almost unrecognizable (fig.5).

Fig.4 Sutherland 12dAX7, left-channel departure from linearity, 16-bit data (2dB/vertical div.).

Fig.5 Sutherland 12dAX7, waveform of undithered 1kHz sinewave at -90.31dBFS, 16-bit data.

As I wrote above, the 12dAX7 needs to see a high input impedance if its dynamic range is not to be compromised. This is confirmed by fig.6, which shows how the unit's THD+noise percentage changes with increased level into three loads: 100k, 10k, and 1k ohms. Into the highest impedance, the actual distortion is beneath the noise level up to 2V output, at which point it starts to emerge. Into 10k ohms, the 12dAX7 clips at 2V output, with the bottom peaks of the waveform squaring off. The Sutherland is obviously in distress driving the admittedly demanding 1k load, with only a few tens of millivolts available.

Fig.6 Sutherland 12dAX7, THD+N (%) vs output voltage into (from left to right): 1k ohm, 10k ohms, 100k ohms.

However, even into a kind 100k ohms and with the volume control used to reduce the analog level to 1V, there was more distortion present than I would have liked to have seen (fig.7). Yes, the second harmonic is just below -60dB (0.1%), with the third at -64dB (0.06%), but there are also many higher-order components visible. The 12dAX7 was also disappointing when it came to high-frequency intermodulation. Again with the analog level reduced to 1V, fig.8 shows that the 1kHz difference component lies at a fairly high -54dB (0.2%). This rose to -40dB (1%) when the volume control was rotated to its maximum position.

Fig.7 Sutherland 12dAX7, spectrum of 50Hz sinewave, DC-1kHz, at 1V into 100k ohms (linear frequency scale).

Note the high noise floor in fig.8, as well as the rise in noise around each of the high-frequency components. This rise is generally an indicator of word-clock jitter problems. When I looked at the 12dAX7's jitter with the Miller Analyzer, I measured a very high 3.44 nanoseconds (3440 picoseconds) peak-peak. The narrowband spectral analysis shown in fig.9 reveals that almost all of this jitter is due to sidebands at ±60Hz (brown "4" numerical markers) and ±240Hz (blue "12"). Actual data-related jitter (red markers) is very low, the sidebands at ±229Hz (red "11") contributing just 93ps of jitter. The jitter didn't change significantly when I drove the 12dAX7 from a high-speed USB 2.0 PCI card instead of the USB 1.0 ports on my computer's motherboard.

Fig.8 Sutherland 12dAX7, HF intermodulation spectrum, DC-24kHz, 19+20kHz at 1V into 100k ohms (linear frequency scale).

Fig.9 Sutherland 12dAX7, high-resolution jitter spectrum of analog output signal (11.025kHz at -6dBFS sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.

It could be argued that the sonic compromises inherent in the low-data-rate MP3-encoded music with which it will be used are far more egregious than the 12dAX7's technical limitations. However, given the engineering excellence of other Ron Sutherland-designed audio components I have had on my test bench, I was disappointed by the 12dAX7. Even if the 60Hz hum problem I experienced was specific to my test conditions—again, MF experienced hum on only one occasion—the general performance of the unit is not very good. In fact, it is worse than implied by Burr-Brown's spec sheets for the PCM2702 chip, particularly regarding noise and output drive ability.—John Atkinson

Sutherland Engineering
P.O. Box 1633
Lawrence, KS 66044
(913) 841-3355