Abbingdon Music Research DP-777 D/A processor Measurements
Sidebar 3: Measurements
I measured the Abbingdon Music Research DP-777 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.
The first thing was to decide which aspects of the DP-777 to measure. Unless I lost count, there are 193 different configurations in which this product can be used, and to perform a full set of measurements on each would take me several weeks. However, as Art Dudley used primarily the USB input and just two configurations, the DP-777's default settings for 44.1kHz and high-resolution data, Classic with the Bit-Perfect II filter and HD with the Organic filter, respectively, from its unbalanced RCA outputs, I concentrated on those modes and spot-checked the other possibilities. I didn't investigate the effects of all the oversampling options, as those are disabled in the two default configurations.
Although Art didn't use the DP-777 as an analog preamplifier, I did check its behavior in this mode. The input impedance of the analog inputs was 37.5k ohms across the audioband and the maximum gain from the variable outputs was 0.1dB; ie, just less than unity gain. The volume control operated in accurate 1dB steps. The analog inputs didn't overload until 5.33V at 1kHz, regardless of the volume-control setting, which is a wide enough margin in practice. The analog inputs preserved absolute polarity and the frequency response in this mode was wide, the output being down by just 0.1dB at 200kHz. Channel separation was superb at >110dB below 2kHz, and distortion was just 0.02% at all frequencies below 1kHz. However, it did rise to 0.2% at 20kHz (fig.1), presumably due to the circuit's decreasing open-loop gain at high frequencies. Pin 2 of the XLR outputs appears to be in parallel with the RCA outputs; this is not a true balanced output. The output impedance at high and middle frequencies was moderately low, at 245 ohms, but at 20Hz rose to 3000 ohms, due to the presence of a DC-blocking capacitor.
Turning to the DP-777's performance as a D/A processor, the maximum output level at 1kHz in the Classic DAC mode was 1.991V; in the HD mode, it was 2.288V. The difference is a very audible 1.2dB in favor of the HD mode, enough to invalidate any A/B comparisons without this difference being compensated for. The maximum outputs with the volume control active were the same, as setting the control to 0dB switches to Bypass mode. All digital inputs preserved absolute polarity (ie, were non-inverting), unless the AMR button on the remote control was held down for several seconds, when it inverted polarity. The S/PDIF, AES/EBU, and USB inputs successfully locked to datastreams with sample rates from 44.1 to 192kHz; the TosLink input was restricted to 96kHz and below. Even with Zero Jitter invoked, I got occasional dropouts with my most jittery source, which was the TosLink input hooked up to the TosLink output of the RME soundcard in one of my test-lab PCs via 15' of plastic TosLink. With my MacBook hooked up to the DP-777, USB Prober identified the AMR as "XMOS USB Audio 2.0" (XMOS seems to be the USB 2.0 solution du jour) operating with 24-bit data in "Isochronous asynchronous" mode, as specified. The DP-777's USB input did successfully operate with sample rates up to 192kHz.
The Filter button on the remote control offers the user two choices of reconstruction filter in Classic mode, four in HD mode. Fig.2 shows the impulse response of Classic's Bit-Perfect I mode. Though this is a perfect pulse, with no ringing or overshoot, it has serious implications for the DAC's frequency-domain behavior (see later, footnote 1). The other Classic filter, Bit-Perfect IIthe default filter in this mode, and used in AD's auditioning of 44.1kHz datawas very similar, other than a slight overshoot the other side of the zero line after the impulse (fig.3). In HD mode, the Traditional filter is a classic time-symmetrical, linear-phase type (fig.4), while the Organic filter (the default in this mode) is close to Bit-Perfect II in Classic mode, but with two extra cycles of ringing after the event (fig.5). The MP Listen filter's impulse response is similar to Organic (not shown), and the Apodizing 808 filter's impulse response shows the familiar ringing after the event, but nothing before it (fig.6): typical minimum-phase behavior.
The blue and red traces in fig.7 show the Classic mode's frequency response with the Bit-Perfect I filter. As well as a slight (0.1dB) level imbalance in favor of the right channel, the output has dropped by just over 3dB at 20kHz. As AMR says that this configuration includes no reconstruction filter of any kind, this would be due to the "aperture effect"; ie, the fact that the pulses comprising the DAC chip's output are finite in length. The Bit-Perfect II filter is claimed to compensate for the aperture effect. As can be seen from the cyan and magenta traces in fig.7, the compensation is effective, though there is still a steep rolloff just below 20kHz.
In HD mode, the Traditional filter follows the familiar picture of an output that starts to roll off gently at the top of each sample rate's passband, before dropping precipitously just below half of each rate (fig.8). Again the right channel is 0.1dB hotter than the left, and the higher sample rates' passbands are a little more restricted than usual, the response with 192kHz data (blue and red data) reaching 3dB at 80kHz. With the Organic filter (fig.9), the output with 44.1kHz data (green and gray traces) rolls off prematurely, reaching 3dB at 18kHz, and the filter's passbands with higher sample rates, while flat in the audioband, are all more restricted than they are with the Traditional filter (fig.10).
Channel separation in Classic mode was poor, at 56dB in both directions below 1kHz, rising slightly to 80dB at 20kHz. (AMR says that this was due to an out-of-spec part used in early production.) It was considerably better in HD mode, at 90dB in both directions below 500Hz, increasing to 100dB LR and 105dB RL at 10kHz.
Since 1989, I have been testing the resolution of digital products by sweeping a 1/3-octave bandpass filter from 20kHz to 20Hz while the device under test decoded dithered data representing a tone at 90dBFS. With the DP-777 in Classic mode and fed 16-bit data, this produced a graph that peaked just below 90dBFS (fig.11), which correlates with a slightly negative linearity error between 90 and 105dB (fig.12). This graph is dominated by analog noise below 90dB, however, suggesting that the Classic mode doesn't quite achieve true 16-bit resolution.
Repeating the resolution test with the DP-777 in HD mode, the top pair of traces in fig.13, taken with dithered 16-bit data, does peak correctly at 90dBFS, and there is no significant linearity error in this mode. However, the noise floor is a little higher than usual with dithered 16-bit data, and odd-order harmonics of the AC-supply frequency can be seen, particularly in the right channel (dotted traces). The middle pair of traces in fig.13 was taken with 24-bit data, but to my surprise the noise floor dropped by just 6dB or so, and then only in the upper midrange and treble, which implies no better than 17-bit resolution in this modenot enough to allow the DP-777 to correctly decode 24-bit data representing a dithered tone at 120dB (bottom traces in fig.12). Both this limited resolution of the HD mode and the presence of odd-order AC-supply harmonics in the right channel were confirmed by FFT analysis (fig.14). This graph was taken via the TosLink input; the result was the same for USB data.
Noise modulation was low, perhaps as a result of the higher-than-expected level of the noise floor with 24-bit data, and this noise did obscure the waveform of an undithered sinewave at exactly 90.31dBFS with both Classic and HD DACs (fig.15). With the HD DAC, increasing the bit depth to 24 gave only a slightly better representation of a sinewave (fig.16), which was to be expected from the relatively limited result on the resolution test.
The DP-777's tubed output stage gave a moderate level of distortion with full-scale digital data, with the second harmonic predominant at around 60dB (0.1%) in both channels (fig.17), dropping to 68dB (0.04%) at 10dBFS (fig.18). However, the DP-777 was not happy driving loads of less than 10k ohms, with high levels of distortion evident (not shown).
With its Traditional filter in HD mode, the DP-777 performed well on the high-frequency intermodulation test, the high-order artifacts all lying below 95dB (fig.19). The second-order difference product at 1kHz lies at 64dB (0.06%). The Apodizing 808 filter behaved similarly, though the Organic filter gave a very different picture (fig.20), with the ultrasonic images of the 19 and 20kHz tones at a very high level and a cluster of aliasing products visible centered on 4.1kHz. The MP Listen filter gave similar results on this test, as did the Bit-Perfect I and II filters in Classic mode. Fig.21 shows the spectrum of Bit-Perfect II: note that the fact that this uses a 16-bit DAC results in a much more granular noise floor with this test compared with the HD DAC's behavior (figs.19 and 20).
Even with its Zero Jitter circuit switched off, the DP-777 offered good rejection of datastream jitter on its AES/EBU and S/PDIF inputs, the main spurious sidebands with the 16-bit J-Test lying at ±47, ±120, ±697, and ±719Hz (fig.22). The 120Hz sidebands are obviously power-supply related, but the others are of unknown origin. The Miller Analyzer estimated the jitter level as 243 picoseconds peakpeak in the left channel, 358ps pp in the right, both of these low in absolute terms. In Zero Jitter mode these figures dropped to 189 and 233ps, respectively; the only jitter-related sidebands left visible are those at ±120Hz (fig.23). The USB input (fig.24) gave slightly higher jitter, though this was still low.
Although the DP-777's Classic mode performed much as I had expected from my experience with other non-oversampling D/A processors, I must admit that I was disappointed with its measured performance in HD mode. I was hoping that the AMR in HD mode would offer a true high-resolution alternative to the NOS behavior, but the former seems compromised. In addition, because the Automatic mode defaults to the 16-bit NOS DAC for 44.1kHz data, playing back 24-bit files sampled at this frequency in this mode will result in those files being truncated to 16 bits.John Atkinson
Footnote 1: The effect of this kind of time-domain behavior on listener preference was one of the things I discussed in my 2011 Richard C. Heyser Memorial Lecture at the 131st Audio Engineering Society Convention, held in New York last October.