Peachtree DACiT 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 Peachtree DACiT (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.
The Peachtree uses the popular Cirrus Logic 8416 S/PDIF receiver chip. The coaxial S/PDIF input 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; this is normal behavior, however, as the TosLink specification doesn't include sample rates above 96kHz. The USB input correctly handled data with sample rates of 32, 44.1, 48, and 96kHz, but not 88.2kHz; the DACiT's Tenor TE7022L USB receiver chip is unable to operate at that sample rate, or at any rate above 96kHz. The Mac's USB Prober utility identified the DACiT as being a "Peachtree USB DAC" from "Peachtree audio," and revealed that the DACiT could operate as a two-channel processor with 16- or 24-bit capability in the normal "isochronous adaptive" mode. I found one anomaly: Not every music player program set the bit depth to 24. I was using Bias Peak Pro to play my test-signal WAV files; when I played 24-bit files, I got obvious truncation at the 16th bit, so I checked AudioMidi Set-Up: the DACiT had defaulted to 16 bits. I believe that Amarra, which Jon Iverson used for the review, correctly handles this, but the Peachtree's bit rate can always be manually set to 24.
The Peachtree's maximum output level at 1kHz was 2.04V, sourced from a usefully low impedance of 250 ohms across the audioband, and it preserved absolute polarity (ie, was non-inverting). Fig.1 shows the DACiT's frequency responses with data sampled at 44.1, 96, and 192kHz; each response follows the same curve, but with a sharp rolloff just below half the same rate. Channel separation (not shown) was asymmetrical: the RL crosstalk was less than 105dB below 5kHz, but the LR leakage was 95dB at all frequencies below 10kHz. Channel separation at 20kHz was >90dB in both directions at 20kHz, which is still excellent.
For reasons of consistency with the reviews I've published since 1989, my first test of a D/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.2. With 16-bit data (top pair of traces), all that can be seen above 1kHz is the spectrum of the dither noise used to encode the signal; below that frequency the DACiT's noise floor starts to intrude, with a very slight bump at the AC line frequency of 60Hz. With 24-bit data (middle pair of traces), the noise floor drops by about 15dB in the treble, implying resolution of between 18 and 19 bits, though the noise floor looks rather granular. At low frequencies, however, the 24-bit data's resolution is obscured by the Peachtree's self noise. A similar picture of the DACiT's resolution can be seen in fig.3, which repeats the spectral analysis with a more insightful FFT technique, though some curious low-level spuriae are apparent with the 24-bit version of the signal at frequencies unrelated to the signal frequency.
This anomalous behavior can also be seen in the bottom pair of traces in fig.2, which show the spectrum of the DACiT's output while it reproduced a dithered 24-bit tone at 120dBFS. While there is a small peak at 1kHz, the traces are disturbed by higher-level peaks at 800Hz and 1.7kHzthe ESS 9023 D/A chip used in the Peachtree appears to be producing "idle tones" with these data. I looked at this behavior in more detail: As you can see in fig.4, a dithered 24-bit tone at 110dBFS was correctly decoded, with no idle tones apparent in this 1/3-octavesmoothed spectrum. It was only when the signal dropped below this level that the D/A chip misbehaved. As real music never has information at this level without higher-level information also being present, this misbehavior will probably have no audible consequences.
Though its noise floor is higher than usual at low frequencies, the Peachtree still managed to correctly reproduce the waveform of an undithered tone at exactly 90.31dBFS with both 16-bit data (fig.5) and 24-bit data (fig.6), and with no significant linearity error until below 110dBFS (not shown), which is good performance. A small degree of noise modulation was apparent, the noise floor rising as the level of a 1kHz tone increased from 60dBFS and 40dBFS (fig.7, green and gray, cyan and magenta traces, respectively) to 0dBFS (blue and red). Spectral peaks can also be seen in this graph at 60Hz and its odd harmonics, though these are all much too low in level to be audible.
When I fed the DACiT data representing a full-scale 50Hz tone. I was surprised to see that the output was actually just starting to clip, with a picket fence of low-level harmonics apparent in the spectrum, and these higher in the right channel (fig.8, red trace) than in the left (blue). This graph was taken into the very kind 100k ohm load; reducing the load to 600 ohms resulted in the Peachtree's output stage being fully clipped (not shown). Reducing the signal level to 10dBFS decreased the distortion harmonics to a vanishingly low level, with the second harmonic dropping from 73dB left and 70dB right to, respectively, 94 and 96dB (fig.9). A similar picture emerged with the high-frequency intermodulation test: dropping the signal level below 0dBFS dramatically reduced the amount and level of intermodulation products (figs. 10 and 11). Though there appears to be a DCDC converter on the Peachtree's circuit board, presumably to increase the incoming 9V from the wall-wart supply, there appears to be insufficient power-supply headroom to cope with full-scale signals without adding some distortion. (Though it is fair to note that this distortion looks worse than it will sound on the spectral plot, due to the fact that other than the third, all the higher harmonics lie at or below 80dB.) This may well not matter with real music, or at least with classical and jazz, in which peaks only occasionally hit 0dBFS. However, it might be a problem with modern pop recordings, which bang their heads against the 0dBFS limit more or less continuously.
Though the DACiT offers good resolution, this is offset to some extent by its jitter rejection. Although it doesn't suffer from data-related jitter, thus not amplifying the level of the high-order harmonics of the 16-bit J-Test signal's low-frequency LSB-level squarewave (fig.12, cyan and magenta traces), some spectral spreading is evident due to the presence of random low-frequency jitter, which also affects the DACiT's reproduction of the 24-bit J-Test signal (fig.12, blue and red). This graph was taken with TosLink data; USB data gave the same result.
Considering its affordable price, the compromises in the Peachtree DACiT's measured behavior seem to be well arranged in that, other than that lack of high-level headroom, they will have no audible consequences. And perhaps that lack of headroom would disappear with an upgraded external power supply?John Atkinson