Bricasti Design M1 D/A converter Measurements
Sidebar 3: Measurements
I used Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system to measure the Bricasti M1 (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 Bricasti's electrical inputs successfully locked to datastreams with sample rates ranging from 44.1 to 192kHz; the TosLink input would not lock to datastreams with sample rates greater than 96kHz, which is normal. As supplied for review, the M1's maximum output level was 4.3V from the balanced XLR jacks and 2.01V from the single-ended RCAs, sourced from output impedances of 58 and 29 ohms, respectively, at all frequencies. Both sets of outputs preserved absolute polarity (ie, were non-inverting), the XLRs being wired with pin 2 hot.
I examined the impulse response of each of the seven filters by feeding the M1 digital black data into which I had inserted a single sample at full scale. The filters all had similar time-symmetrical, linear-phase impulse responses; fig.1 was taken with Filter 4. However, the filters did differ in their frequency-domain behavior. Fig.2 shows the response of Filter 1 with data sampled at 44.1kHz (green and gray traces), 96kHz (cyan, magenta), and 192kHz (blue, red). This filter has a sharp rolloff just below 20kHz with 44.1kHz data, though the higher sample rates follow the gentle rolloff seen above 10kHz. Both 96 and 192kHz roll off earlier than with other D/A processors, the responses being 3dB at 33 and 55kHz, respectively. Fig.3 shows the behavior of Filter 4. The 96 and 192kHz responses are identical to Filter 1, but some passband ripple is now evident at 44.1kHz. This is generally felt not to be a good thing, but John Marks actually preferred the sound of Filter 4 to the other six. Fig.4 shows the response of Filter 6, which I preferred, while fig.5 shows the response of Filter 0, which neither of us liked but which gives the widest bandwidth at all sample rates.
The M1's channel separation (not shown) was superb, at >125dB in both directions below 1kHz and still 113dB at the top of the audioband. For reasons of consistency with the digital tests I have performed since 1989, my first test of a processor's dynamic range is to sweep a 1/3-octave bandpass filter from 20kHz to 20Hz while the processor decodes a dithered 1kHz tone at 90dBFS. The results of this test are shown in fig.6: with 16-bit data (top pair of traces), all that can be seen is the spectrum of the dither noise used to encode the signal. With 24-bit data (middle pair of traces), the noise floor drops by 20dB, implying that the M1 has almost 20 bits' worth of dynamic range, easily enough to allow the decoding of a dithered tone at 120dBFS (bottom traces). This is excellent performance, and, just as important, the lowering of the noise floor with the greater bit depth has not unmasked any supply-related spuriae. FFT analysis confirms this excellent resolution (fig.7), and no harmonic distortion components can be seen, though a supply component at 180Hz is now evident in the left channel at a roots-of-the-universe 137dBFS!
Looking at how the Bricasti's noise floor changed with differences in signal level, nothing was evident other than what could be attributed to the Audio Precision's gain-ranging circuitry (fig.8). Similarly, all that could be seen in the graph of the M1's linearity error was the recorded dither noise, so I haven't shown it. With its very low background noise and excellent linearity, the Bricasti's reproduction of an undithered tone at exactly 90.31dBFS was superbly symmetrical; the three DC voltage levels and the Gibbs Phenomenon "ringing" on the waveform's leading edges were all cleanly defined (fig.9, footnote 1). With undithered 24-bit data, the result was a good representation of a sinewave, despite the very low signal level (fig.10).
Harmonic distortion into high impedances was very low, and predominantly the third and fifth harmonics (fig.11), these respectively lying at 110dB (0.0003%) and 119dBFS (0.0001%). These odd-order harmonics rose by 10dB into the punishing 600 ohm load (fig.12), and were joined by the second and fourth harmonicsbut in absolute terms, they are all still very low in level. The Bricasti's performance in the high-level, high-frequency intermodulation test depended on which reconstruction filter was selected. The levels of the difference tone at 1kHz and the higher-order intermodulation products at 18 and 21kHz were the same with all filters and were all very low in level, but the best rejection of ultrasonic images of the 19 and 20kHz tones was with Filter 0 (fig.13), the worst with Filter 4 (fig.14). Although a couple of aliasing products are visible in the audioband with Filters 4 and 6, these are still at 130dB or lower and are therefore inconsequential.
The Bricasti M1's rejection of jitter was one of the best I have measured; any jitter-related spuriae lay below the resolution limit of the Miller Analyzer. The cyan and magenta traces in fig.15 show the spectrum of the M1's analog output while it decoded a 16-bit version of the diagnostic J-Test signal via its TosLink input. The spectral lines visible are the residual odd-order harmonics of the low-frequency squarewave; these are not accentuated in any way, nor are any other sidebands visible other than a single pair at ±180Hz, these lying at almost 140dB. With the 24-bit version of the J-Test (blue and red traces), all the squarewave harmonics have disappeared, and the central spike that represents the 11.025kHz tones has narrower skirts. Jitter rejection was just as impressive via the M1's AES/EBU and S/PDIF inputs.
Bricasti Design's M1 has state-of-the-art measured performance.John Atkinson
Footnote 1: The significance of this test is that in the 2s-complement encoding used in the Compact Disc, the transition from 0 to +1LSB involves just the LSB changing value, while the transition from 0 to 1LSB involves all 16 bits changing value. It therefore offers a quick way of identifying bit-magnitude errors (though with modern delta-sigma D/A converters, such errors are very rare).