Esoteric D-07 D/A processor Measurements
Sidebar 3: Measurements
I used Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system to perform the measurements on the Esoteric D-07 (see www.ap.com and the January 2008 "As We See It"); for some tests, I also used my Audio Precision System One Dual Domain and the Miller Audio Research Jitter Analyzer. To test the D-07's performance via its USB input, I used my MacBook running OS10.6 and Bias Peak Pro 6 to play WAV files of the various test signals I use.
Apple's USB Prober utility identified the D-07 as being manufactured by "ESOTERIC USB Audio" and indicated that it would accept 16- and 24-bit data with sample rates of 32, 44.1, 48, and 96kHz. Missing in action was the 88.2kHz sample rate; according to Esoteric, when the design of the D-07 was finalized, the Tenor/USB chip set used did not support native 88.2kHz data via USB. The USB Prober utility confirmed that the Esoteric's USB input operates in "isochronous adaptive" mode, in which the host computer controls the flow of data to the DAC.
All measurements were taken with the digital volume control set to its maximum. The maximum output level at 1kHz from both the balanced and unbalanced jacks was 2.16V, or 0.65dB above the CD standard's 2V RMS. Both outputs were non-inverting (ie, preserved absolute polarity), and the output impedance was to specification at high and midrange frequencies, at 200 ohms balanced and 100 ohms unbalanced. Both figures rose inconsequentially, to 218 and 106 ohms, respectively, in the low bass. Channel separation (not shown) was superb, at better than 120dB below 2kHz.
The D-07 offers a choice of two reconstruction filters, labeled FIR and S_DLY, which have identical frequency responses at all sample rates. Fig.1 shows the response of the FIR setting at rates of 44.1, 96, and 192kHz; the top-octave output rolls off gently at all three rates, to reach 0.3dB at 20kHz, with then a steep falloff just before half each sampling frequency. With 192kHz data, the response drops to 3dB at 65kHz. Fig.2 shows the impulse response of the linear-phase FIR filter, which, as expected, is symmetrical in the time domain. By contrast, fig.3 shows the impulse response of the S_DLY filter; this is a minimum-phase type, with all ringing occurring after the event.
Fig.1 Esoteric D-07, frequency response at 12dBFS into 100k ohms from balanced outputs with data sampled at: 44.1kHz (left channel green, right gray), 96kHz (left cyan, right magenta), 192kHz (left blue, right red; 0.25dB/vertical div.).
Fig.2 Esoteric D-07 set to FIR Filter, response to single sample at 0dBFS, 44.1kHz-sampled data (4ms time window).
Fig.3 Esoteric D-07 set to S_DLY, response to single sample at 0dBFS, 44.1kHz-sampled data (4ms time window).
I tested the Esoteric's resolution in my usual fashion, feeding it 16- and 24-bit data representing a dithered 1kHz tone at 90dBFS while sweeping the center frequency of a 1/3-octave bandpass filter from 20kHz to 20Hz; the resulting spectra, taken without oversampling, are shown in fig.4. The traces peak at exactly 90dBFS, suggesting very low linearity error, with no distortion- or power-supplyrelated spuriae visible for both 16- and 24-bit data. The increase in bit depth drops the noise floor by around 12dB, implying at least 18-bit resolution, which is confirmed by the FFT spectra in fig.5. Switching in oversampling to DSD produced the top pair of traces above 10kHz in fig.4; what should be a smooth rise in the noise floor actually looks slightly scalloped. I have no idea what this is due to or what it means. Changing to dithered 24-bit data representing a 1kHz tone at 120dBFS without upsampling gave the bottom pair of traces in fig.4. The tone was easily resolved with both AES/EBU data and USB 24-bit data.
Fig.4 Esoteric D-07, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (top) and 24-bit data (middle), and of dithered tone at 120dBFS with 24-bit data (right channel dashed).
Fig.5 Esoteric D-07, 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).
With its excellent DAC linearity (fig.6) and low levels of analog noise, it was no surprise that the D-07 reproduced an undithered 16-bit tone at exactly 90.31dBFS with excellent waveform symmetry, or that the three DC voltage levels described by these data readily resolved (fig.7). Changing to undithered 24-bit data gave a well-formed sinewave (fig.8).
Fig.6 Esoteric D-07, linearity error, dBr vs dBFS, 16-bit data (2dB/vertical div.)
Fig.7 Esoteric D-07, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data (left channel blue, right red).
Fig.8 Esoteric D-07, waveform of undithered 1kHz sinewave at 90.31dBFS, 24-bit data (left channel blue, right red).
All the remaining graphs were taken with the FIR filter; the S_DLY filter didn't give different results. Fig.9 shows the spectrum of a full-scale 1kHz tone into 100k ohms. The third and fifth harmonics are the highest in level, but I can safely say that, at 106dB (0.0005%), these will not be audible; more important, they didn't increase in level when I switched to a 600-ohm load. I was more concerned by the spreading of the spectral peak representing the 1kHz tone and the presence of sidebands at ±120 and ±240Hz. (Experimenting with the grounding between the D-07 and the SYS2722 didn't change this behavior.) This spectrum was taken without upsampling; upsampling to DSD produced a similar distortion signature, but now with a granular-looking noise floor (fig.10).
Fig.9 Esoteric D-07, spectrum of 1kHz sinewave, DC10kHz, at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.10 Esoteric D-07, spectrum of 1kHz sinewave, DC10kHz, at 0dBFS into 100k ohms, upsampled to DSD (left channel blue, right red; linear frequency scale).
I investigated the spectral spreading by repeating the analysis with 1kHz tones at 20 and 60dBFS, but restricting the bandwidth of the graph to 1kHz. The result is shown in fig.11: the blue and red traces are the spectrum of the tone at 0dBFS, the green and gray traces with the tone at 20dBFS, the cyan and magenta traces with the 60dBFS tone. You can see that both the spectral spreading and the presence of the supply-related sidebands are related to signal level. They are at their maximum with the full-scale tone, but have disappeared with the tone at 60dBFS. This misbehavior is at a very low level, but the ear does tend to be more sensitive to this kind of noise modulation than you might expect.
Fig.11 Esoteric D-07, spectrum of 1kHz sinewave, DC1kHz, at: 0dBFS into 100k ohms (left channel blue, right red), 20dBFS (left green, right, gray), 60dBFS (left cyan, right magenta). (Linear frequency scale.)
This spectral spreading can also be seen in fig.12, which shows the spectrum of the D-07's output while it decoded 24-bit data representing a full-scale mix of 19 and 20kHz tones without upsampling. Commendably, actual intermodulation products are vanishingly low in level, with the 1kHz difference product at 130dB (0.00003%), and then only in the left channel. Neither reducing the load impedance to 600 ohms nor upsampling to DSD changed the excellent intermodulation performance, though the latter gave the same granular noise floor seen in earlier graphs (fig.13).
Fig.12 Esoteric D-07, HF intermodulation spectrum, DC24kHz, 19+20kHz at 0dBFS into 100k ohms, no upsampling (linear frequency scale).
Fig.13 Esoteric D-07, HF intermodulation spectrum, DC24kHz, 19+20kHz at 0dBFS into 100k ohms, upsampled to DSD (linear frequency scale).
I assessed the Esoteric D-07's rejection of word-clock jitter in many different configurations: TosLink, AES/EBU, or USB inputs; in single- or dual-PLL receiver modes; and with 2Fs, 4Fs, or DSD upsampling. Upsampling to DSD always gave the lowest measured jitter level, with AES/EBU better than TosLink and USB about the same as TosLink. Surprisingly, the dual-PLL mode gave only a small reduction in jitter sidebands, though it did affect the spectrum (see later).
Fig.14 shows the D-07's output spectrum decoding a 16-bit J-Test signal from the Audio Precision SYS2722 via 15' of plastic TosLink cable, with no upsampling and PLL1. Strong sidebands can be seen at ±120Hz (power-supplyrelated) and ±229.5Hz (data-related; all other data-related sidebands are at the test signal's residual level). The Miller Analyzer calculated the level of the sidebands to be equivalent to a high 1340 picoseconds (ps) peakpeak of jitter, which improved to 1065ps pp with PLL2. Upsampling to 2Fs reduced the jitter further, to 900ps; 4Fs gave a slightly worse performance, 985ps, and DSD gave a reduction to 600ps. A similar trend could be seen with AES/EBU data: 1340ps with no upsampling, 705ps with 2Fs upsampling, 695ps with 4Fs, and 646ps with DSD.
Fig.14 Esoteric D-07, 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 TosLink, no upsampling, PLL1 (left channel blue, right red); 24-bit data via AES/EBU (left cyan, right magenta). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.15 shows that switching in the second PLL changes the level of the data-related sidebands but not the supply-related spuriae. But it also adds a broad hump of low-frequency random jitter, which spreads and increases the level of the noise floor either side of the central spike that represents the 11.025kHz tone. In this, the D-07's PLL2 mode very much resembles the Meridian 518 processor I reviewed in January 1996 (see the grayed-out trace in fig.1).
Fig.15 Esoteric D-07, 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 TosLink, no upsampling, PLL2 (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Finally, fig.16 shows the spectrum of the D-07's output while it decoded 16-bit J-Test data via USB without upsampling. The jitter level was 941ps pp, which is quite high, and as well as the familiar sidebands at ±120 and ±229.5Hz, there are strong sidebands of unknown origin at ±76Hz.
Fig.16 Esoteric D-07, 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, no upsampling (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
In most ways, the Esoteric D-07 offered excellent measured performance. But I was disappointed in its relatively poor rejection of jitter and the noise modulation with high-level signals.John Atkinson