Benchmark DAC1 USB D/A processor & headphone amplifier Follow-Up, July 2008
When, in January, I reviewed this superb-sounding and relatively affordable D/A processor ($1275), a couple of aspects of its behavior left me puzzled. Fed with a test tone, for example, my sample appeared to go deaf below 67dBFS when driven with 16-bit files by my laptop via the Benchmark's USB connection. I listened to the 500Hz fade-to-noise-with-dither track from the CBS Test CD, ripped to my PowerBook, and the tone vanished about 5 seconds after the start of the fade. This didn't happen when I mixed a high-level 19kHz tone with the test signal, nor did it happen with 24-bit data. Needless to say, I was left very puzzled.
It turned out that the original sample included an undocumented feature: when the level of 16-bit audio data dropped below about 70dBFS, the DAC1 muted its output (see Benchmark's "Manufacturer's Comment," January, p.157; www.stereophile.com/digitalprocessors/108bench/index5.html). Benchmark pointed out that, unlike test signals, real-world music is very unlikely to have zero DC offset, no noise, and no high-frequency content. However, they revised the DAC1 USB's firmware to eliminate this feature and sent me a new sample representative of current production (footnote 2).
The second sample, driven both from its AES/EBU input and from its USB input (using the Mac mini that acts as our household's media server), sounded identical to the original DAC1 USB: balanced a little forward, but with superbly transparent soundstaging, clean high frequencies, and powerful lows.
The fade-to-noise test tone sounded as it should, without dropped-out data, so I ran some quick tests using my older Audio Precision System One Dual Domain test system. The USB data source was my Macintosh G4 TiBook (OSX 10.3.9, USB 1.1). To avoid the danger of having iTunes inadvertently do something "helpful," I played back AIF and WAV files using the inexpensive Amadeus II audio workstation program.
Fig.1 Benchmark DAC1 USB, USB data connection, left-channel departure from linearity, 16-bit data (2dB/vertical div.).
Fig.1 shows the result of the fade-to-noise tests, using a dithered 16-bit, 500Hz tone that drops in level from 60dBFS to 120dBFS. The tone drops linearly, without any level error, to below 100dBFS, when random noise increasingly leads to a positive error. This noise is actually the recorded ditherthe Benchmark's noise floor is lower than this.
Fig.2 Benchmark DAC1 USB, USB data connection, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit (top) and 24-bit (bottom) data (right channel dashed).
The same thing was true when I performed a spectral analysis of the DAC1's analog output while it decoded dithered 16-bit data representing a 1kHz tone at 90dBFS, using a 1/3-octave bandpass filter swept from 20kHz down to 20Hz (fig.2, top pair of traces). The left- and right-channel traces peak at exactly 90dBFS, no harmonic or power-supply spuriae are apparent, and the noise floor is dominated by the recorded dither. Increasing the bit depth to 24 gave the lower traces in fig.2; the peak still correctly touches the 90dB line, but now the noise floor is 17dB lower across the band (footnote 2). This test result confirms that the Benchmark's USB input does correctly handle 24-bit data.
Fig.3 Benchmark DAC1 USB, USB data connection, frequency response at 12dBFS into 100k ohms, 96kHz data (right channel dashed, 1dB/vertical div.).
The DAC1 USB also correctly handles highsample-rate data: the frequency-response traces in fig.3 were taken with 24-bit data sampled at 96kHz and fed to the DAC1's USB input. The response was flat within the audioband, and gently rolled off by 0.8dB at 40kHz, before beginning the expected precipitous decline above 42kHz.
The circuit topology used by Benchmark endows the DAC1 with excellent immunity to word-clock jitter, which was confirmed by my tests of the original sample. The Miller Audio Research Jitter Analyzer measured a very low 157 picoseconds peakpeak of word-clock jitter with the DAC1 fed a 16-bit version of the Miller/Dunn J-Test signal via its TosLink S/PDIF data input, while increasing the word length to 24 bits dropped the measured jitter level to 119ps pp. Fed 16-bit data via its USB input, the second sample of the DAC1 USB gave a slightly higher jitter level of 296ps peakpeak. This was almost entirely due to a pair of sidebands of unknown origin at ±3194Hz; data-related sidebands were at the residual level.
The modification to the DAC1 USB's firmware solved the dropout mystery. Not only did the second sample measure as well as I'd expected it to, the measurements also validated Benchmark's approach to implementing the DAC1's USB input, I feel. This really is a nicely engineered piece of gear.John Atkinson
Footnote 1: The serial number of the original sample was 00022, that of the new sample 0722323.
Footnote 2: The 24-bit noise floor was 34dB lower when I performed this test on the original sample. This was because I had set the DAC1 USB's fixed output level to its maximum. For the tests on the second sample, I used it with the output padded down by 10dB.