Rega DAC D/A processor Measurements
Sidebar 2: Measurements
In his May 2011 column, Sam Tellig enthused about the sound of the new $995 DAC from UK-based Rega Research. "The Rega DAC had a richness, a fullness of tone, an analog sense of ease, that I had not hitherto heard from digital, save for SACD," Sam wrote. "I heard a naturalness, an organic quality to the sound. . . . I enjoyed that richness of tone: body, weight, authority."
Jon Iverson listened to the Rega DAC while preparing his review of the Peachtree Audio iDac in the October 2011 issue. While Jon agreed with Sam about the Rega's low frequencies, he wasn't convinced that this was a benefit. Compared with the Peachtree, "the Rega had a slightly heavier feel in the bottom end and added girth to voices (I was surprised at how noticeable this bass boost was)," he wrote, while his ex-audio retailer listening buddy Bruce Rowley added that, against Jon's long-term reference, the Benchmark DAC 1, "the Rega [sounded] warmer but a bit fuzzy. . . . the Rega adds body to vocals."
It seems obvious that both Sam and Jon were describing the same character but differed in their value judgments, something that might well have been system dependent. (This is why Stereophile's reviewers always list their review systems.)
There was also some debate about the Rega DAC's USB input. As Sam wrote, this uses the Burr-Brown PCM2707 USB receiver chip, which is limited to data rates of 48kHz and below and word lengths of 16 bits. According to the Rega's designer, Terry Bateman, the DAC's PCM2707 reclocks the incoming datastream "on the fly," in order to avoid data loss and reduce jitter. I had assumed that by this Bateman was referring to the statement, in Burr-Brown's datasheet for the PCM2707, that it "employs SpAct architecture, [Texas Instruments'] unique system that recovers the audio clock from USB packet data." Sam reported Bateman as saying that this is what would be achieved by DACs that use an asynchronous USB connection.
Ayre Acoustics' Charles Hansen took issue with that statement. In a letter published in the July 2011 Stereophile (p.133), he wrote that the Rega DAC's USB receiver operates in what is called "adaptive" mode, in which the DAC is completely controlled by the computer. The optimal means of operating an audio USB interface, wrote Hansen, is called "asynchronous," in which "timing errors in the incoming datastream are absorbed by a small buffer and the audio data are clocked out by a local, fixed-frequency master clock." In an asynchronous USB receiver, Hansen continued, "the DAC is in charge of the entire system, and the computer is slaved to it." By contrast, "the master audio clock in an adaptive USB DAC must be variable in frequency. (This allows the unit to adapt to the timing variations in each computer.) But if all else is equal, a variable-frequency clock will always have higher levels of jitter than a fixed-frequency clock."
I asked Rega's US distributor to send me a sample of the Rega DAC so I could investigate this matter for myself. I also subjected this sample (serial no.01631) to my usual set of tests, using Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see www.ap.com and the January 2008 "As We See It"); for some tests, I also used my vintage Audio Precision System One Dual Domain and the Miller Audio Research Jitter Analyzer. As well as driving the Rega DAC with S/PDIF data from the Audio Precision analyzers, I used a MacBook running Mac OS10.6.8 and Pure Music 1.8 to play test-signal files.
Before doing any testing, I took a look under the hood to confirm that the Rega DAC does use the Burr-Brown PCM2707; it does. For its S/PDIF inputs it uses a Wolfson WM8805 transceiver chip, which will handle data with sample rates of up to 192kHz. This feeds two Wolfson WM8742 D/A chips, a 24-bit/192kHz part; the pair of chips can be seen in the photo of the Rega's innards, just above the legend "Best Used With EL84 Valves" on the printed circuit board. (The EL84, aka the 6BQ5, is a small power pentode tube made famous by its use in the classic Vox AC30 guitar amplifier.)
Via its S/PDIF inputs, the Rega DAC successfully locked to data with sample rates of up to 192kHz. What I found most interesting about the Rega DAC is that it offers 10 different digital filters, selected with a pushbutton on the front panel. At sample rates up to 48kHz, these filters are described in the owner's manual as: 1) Linear-phase half-band filter, 2) Minimum-phase soft-knee filter, 3) Minimum-phase half-band filter, 4) Linear-phase apodizing filter, and 5) Minimum-phase apodizing filter. At higher sample rates, the filters offered are: 1) Linear-phase soft-knee filter, 2) Minimum-phase soft-knee filter, 3) Linear-phase brickwall filter, 4) Minimum-phase apodizing filter, and 5) Linear-phase apodizing filter.
Sam didn't try the filters with recordings of data rates higher than 44.1kHz, but wrote that, "while Rega recommends using Filter Setting 1 as the default, you might find that Filters 4 and 5 have their charms." The apodizing filters should offer quicker, cleaner, clearer transients, and that's exactly what Sam heard, especially from Filter 4, though he did add that the filters' effects are subtle. Fig.1 shows the impulse response of Filter 4 at 44.1kHz. It is indeed a linear-phase filter, as evidenced by the fact that the ringing occurs symmetrically before and after the transient peak. I'm not sure why, therefore, Rega refers to this as an "apodizing" filter, which, as originally defined by Peter Craven, has a null at the Nyquist Frequency (half the sample rate) and therefore eliminates the pre-ringing. Perhaps Rega means that Filter 4's ringing is lower in frequency than the original data's Nyquist frequency of 22.05kHz, and that it still has a null at that frequency.
Fig.1 Rega DAC, Filter 4, response to single sample at 0dBFS, 44.1kHz-sampled data (4ms time window).
By contrast, fig.2 shows Filter 5's impulse response; this is a minimum-phase type, as specified, with all the ringing following the transient. Filters 2 and 3 have impulse responses that appear identical to that of Filter 5. Filter 1 has an impulse response very similar to that of Filter 4. As can be seen from these impulse responses, the Rega DAC's output preserved absolute polarity (ie, was non-inverting). The Rega's maximum output level was not affected by the choice of filter; it measured 2.17V at 1kHz, sourced from a fairly low impedance of 620 ohms.
Fig.2 Rega DAC, Filter 5, response to single sample at 0dBFS, 44.1kHz-sampled data (4ms time window).
Turning to the frequency domain, Filter 3 gave the slowest rolloff at all sample rates (fig.3), Filters 4 and 5 the earliest rolloff (fig.4), as you'd expect from their apodizing nature. All the filters offered superb ultrasonic image rejection, however. Channel separation at 1kHz was also superb at 106dB, RL, and 109dB, LR, decreasing to 80 and 85dB at 20kHz, respectively, due to the inevitable capacitive coupling between channels.
Fig.3 Rega DAC, Filter 3, frequency response at 12dBFS into 100k ohms with data sampled at: 192kHz (left channel green, right gray), 96kHz (left cyan, right magenta), 44.1kHz (left blue, right red). (0.25dB/vertical div.)
Fig.4 Rega DAC, Filter 5, frequency response at 12dBFS into 100k ohms with data sampled at: 192kHz (left channel green, right gray), 96kHz (left cyan, right magenta), 44.1kHz (left blue, right red). (0.25dB/vertical div.)
The top pair of traces in fig.5 shows a 1/3-octave spectral analysis of the Rega's outputs while it decoded dithered 16-bit S/PDIF data representing a 1kHz tone at 90dBFS. The traces peak at 90dBFS, suggesting very low linearity error (not shown), and while the left channel's noise floor (solid trace) is dominated by the dither noise, the right channel's (dashed trace) has a peak at 60Hz, implying some magnetic coupling from the power transformer. At 110dBFS, however, this is not going to be audible. Increasing the bit depth to 24 (middle traces) lowered the noise floor by 10dB, implying almost 18-bit resolution, though this did unmask a slight amount of supply-related 120Hz content in both channels at a negligible 126dB, confirmed by FFT analysis (fig.6). This is sufficient resolution to allow the Rega to reproduce a dithered tone at 120dBFS (fig.5, bottom traces). The Rega correctly represented an undithered sinewave at 90.31dBFS with 16-bit data (fig.7) and 24-bit data (fig.8).
Fig.5 Rega DAC, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS, with: 16-bit data (top), 24-bit data (middle), dithered 1kHz tone at 120dBFS with 24-bit data (bottom). (Right channel dashed.)
Fig.6 Rega DAC, left-channel linearity error, 16-bit data (dBr vs dBFS, 2dB/vertical div.).
Fig.7 Rega DAC, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS, with: 16-bit data (left channel cyan, right magenta), 24-bit data (left blue, right red).
Fig.8 Rega DAC, Filter 1, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data (left channel blue, right red).
The power-supplyrelated spuriae can also be seen in fig.9, which shows how the Rega's noise floor varies in level as the signal level changes. Though the variation is small, what did concern me was the appearance of sidebands at ±120Hz with a signal at 0dBFS. Only the lower sideband is shown in this graph; extending the measurement bandwidth to 10kHz shows a clearer picture of the sidebands (fig.10). This graph was taken into the demanding 600 ohm load; a regular series of harmonics can be seen, with the second and third the highest in level, at 86dB (0.005%) and 89dB (0.0003%), respectively. Increasing the loading to the benign 100k ohms didn't change the levels of the harmonics, other than to reduce the second harmonic to 90dB (not shown). The Rega DAC offered very low levels of intermodulation distortion, with the only differences between the filters affecting how much ultrasonic content was present. Fig.11 shows the spectrum with 44.1kHz data and Filter 1, fig.12 with 44.1kHz data and Filter 5, which were the two extremes. None of the filters featured the poor image rejection typical of non-oversampling DACs, or the slow-rolloff filters featured in some other DACs.
Fig.9 Rega DAC, Filter 1, waveform of undithered 1kHz sinewave at 90.31dBFS, 24-bit data (left channel blue, right red).
Fig.10 Rega DAC, spectrum of 1kHz sinewave, DC1kHz, at: 0dBFS (left channel blue, right red), 40dBFS (left cyan, right magenta), 60dBFS (left blue, right red). (Linear frequency scale.)
Fig.11 Rega DAC, spectrum of 1kHz sinewave, DC1kHz, at 0dBFS into 600 ohms (left channel blue, right red; linear frequency scale).
Fig.12 Rega DAC, Filter 1, HF intermodulation spectrum, DC30kHz, 19+20kHz at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
The Rega DAC's jitter rejection via its S/PDIF input was good with respect to data-related sidebands, which lie at the residual level with 16-bit J-test data (fig.13, cyan and magenta traces), and are absent with 24-bit J-Test data (blue and red traces). However, pairs of sidebands can be seen at the supply-related frequencies of ±120 and ±240Hz, with the Miller Analyzer estimating their level as equivalent to 273 picoseconds peakpeak.
Fig.13 Rega DAC, Filter 5, HF intermodulation spectrum, DC30kHz, 19+20kHz at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Turning to the Rega's USB input, Apple's USB Prober utility identified the device as a "USB Audio DAC" by "Burr-Brown from TI," and confirmed that it operated in "Isochronous adaptive" mode. Bit resolution was listed as "16," and the sample rates supported were 32, 44.1, and 48kHz. The Rega's performance via its USB input was identical to that from the S/PDIF inputs, other than the restricted bit depth and sample rate. This was true also of the DAC's jitter spectrum with USB data (fig.14), which was marred by the same supply-related sidebands. However, sidebands at the adaptive-polling intervalrelated frequency of ±1kHz are absent in this graph. In addition, the central spike that represents the 11.025kHz tone with USB data in this graph is as well defined and narrow as it is with S/PDIF data, suggesting that the large amount of the random low-frequency jitter that is often evident with adaptive USB products is also absent. The Rega DAC's USB input is well sorted, as its English designer would say.
Fig.14 Rega DAC, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz: 16-bit data via 15' TosLink S/PDIF from AP SYS2722 (left channel cyan, right magenta), 24-bit data (left blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.15 Rega DAC, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz: 16-bit data via USB from MacBook (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Overall, the Rega DAC offered respectable measured performance for $995. But I am concerned about the ±120Hz sidebands that accompany high-level tones. Psychoacoustic masking theory suggests that these sidebands will be too close in frequency to the musical tones to be perceived. On the other hand, Paul Miller has written, in the English magazine Hi-Fi News & Record Review, that he has found that the presence of low-frequency sidebands in other processors correlates with the perception that a product's bass is larger than life, even a touch ponderous. Although my measurements show that the Rega DAC's frequency response is flat to below 20Hz, both Jon and Sam described the Rega's low frequencies as being exaggerated, which might correlate with this measured problem. Otherwise, it remains a puzzle.John Atkinson