Apogee Electronics PSX-100 digital converter Measurements
In the main, the Apogee PSX-100's D/A section was tested using the digital AES/EBU digital output of an Audio Precision System One Dual Domain. I also made use of a two-card PrismSound DScope v2.02 and a Miller Audio Research Jitter Analyzer, which runs on a National Instruments DSP PC card.
Looking first at the Apogee's D/A section, with its output set to 10dBV (which is how MF used the converter), the front-panel multiturn pots can still be adjusted to give a maximum output level of 7.3V. With the pots set to give a standard CD output level of 2V RMS, switching the rear-panel DIP switch to +4dBU also gave a maximum level of 7.3V RMS. The source impedance was extremely low: less than 1 ohm. However, the unit would not drive 600 ohms at its maximum output, the highest level into this load before clipping being around 5.5V, which should be plenty high enough. A rear-panel DIP switch allows you to define pin 3 or pin 2 of the XLR jacks as positive. With either setting, the D/A output inverted polarity with the Audio Precision's pin-2positive input.
Fig.1 shows the D/A frequency response with a 44.1kHz sample rate, at 12dBFS. The lower traces in the treble are without pre- and de-emphasis, and are impressively flat. However, the upper pair of traces are with pre-emphasized data, and reveal that the Apogee does not apply the correct de-emphasis with such data. On the very small number of CDs that have been recorded with pre-emphasis, the tonal balance will be audibly skewed toward the highs.
Fig.1 Apogee PSX-100, 44.1kHz sampling, D/A frequency response with (top) and without (bottom) de-emphasis at 12dBFS (right channel dashed, 1dB/vertical div.).
Fig.2 shows the D/A response at 0dBFS with 48kHz and 96kHz sample rates. (The source was the Apogee's A/D section converting the Audio Precision's analog output.) At the higher sample rate you can see a gentle rolloff above the audioband, before the plunge at Fs/2, which will correlate with good time-domain behavior. Channel separation (not shown) was superb, at better than 110dB across the band, with only a slight rise in the top audio octave.
Fig.2 Apogee PSX-100, D/A frequency response at 0dBFS at 48kHz and 96kHz sampling frequencies (right channel dashed, 5dB/vertical div.).
Fig.3 shows a 1/3-octave spectral analysis of the D/A's output while it processed data representing a dithered 1kHz tone at 90dBFS. The top traces were taken with 16-bit data, the bottom traces with 24-bit data. The increase in word length gives about a 10dB reduction of the noise floor, suggesting an ultimate noise-limited dynamic range of closer to 18 bits, which is still excellent. (Only a very few of the D/A converters I have measured have achieved better dynamic range than this, and these are still limited to around 20 bits of dynamic range when processing 24-bit data.) Replacing the 1kHz tone with digital black and extending the measurement bandwidth to 200kHz gave the curves shown in fig.4. The right channel has a slight amount of energy apparent at the sampling frequency, but the traces are otherwise free from spuriae.
Fig.3 Apogee PSX-100 D/A, 1/3-octave spectrum of dithered 1kHz tone at 90dBFS, with noise and spuriae, 16-bit (top) and 24-bit (bottom) data (right channel dashed).
Fig.4 Apogee PSX-100 D/A, 1/3-octave spectrum of digital black, with noise and spuriae, 16-bit (top) and 24-bit (bottom) data (right channel dashed).
The PSX-100 offered superb D/A linearity, as can be seen in fig.5. The amplitude error remains below 2dB down to around 114dBFS. This graph was taken with dithered 16-bit data; extending the word length reduced the level at which the amplitude error reached 2dB to 120dBFS.
Fig.5 Apogee PSX-100, D/A departure from linearity, 16-bit data (right channel dashed, 2dB/vertical div.).
The excellent linearity and low noise resulted in an accurate reproduction of the waveform of an undithered 1kHz tone at 90.31dBFS (fig.6). (The data representing this signal consist of just 1LSB, 0, and +1LSB.) The three discrete voltage levels can be easily perceived in this graph, along with the Gibbs Phenomenon "ringing" at the bit transitions. Extending the undithered word length to 24 bits gave the waveform shown in fig.7quite a good facsimile of a sinewavem at the very low signal level.
Fig.6 Apogee PSX-100 D/A, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data.
Fig.7 Apogee PSX-100 D/A, waveform of undithered 1kHz sinewave at 90.31dBFS, 24-bit data.
The Apogee's D/A section was superbly linear, a full-scale 50Hz sinewave output into 100k ohms producing just one distortion harmonic above the analyzer's noise floor: the benign second harmonic, at an even more benign 94dB (0.002%). Dropping the test load to 600 ohms led to clipping at the PSX-100's maximum output level of 7.3V RMS; reducing the output level to 5.5V (fig.8) resulted in a second harmonic at 84dB (0.006%), which is still excellent, though the third, fourth, and fifth harmonics can now be seen at lower levels. High-frequency intermodulation (fig.9) was also superbly low in level.
Fig.8 Apogee PSX-100 D/A, spectrum of 50Hz sinewave, DC1kHz, at 2.5dBFS into 600 ohms (linear frequency scale).
Fig.9 Apogee PSX-100 D/A, HF intermodulation spectrum, DC22kHz, 19+20kHz at 0dBFS into 100k ohms (linear frequency scale).
Only when I examined the effects of word-clock jitter in the PSX-100 D/A's analog output signal did this superb measured performance stumble. I use a 16-bit analytical signal developed by Julian Dunn when he was at PrismSound, and implemented by Paul Miller. It consists of an 11.025kHz tone (44.1kHz/4) at 6dBFS, over which has been overlaid the 16th, or least-significant bit (LSB), toggling on and off at 229Hz. The Miller Jitter Analyzer averages sixty-four 32,768-point FFTs on the processor's analog output while it decodes these data, and looks for symmetrical pairs of sidebands on either side of the high-frequency tone. An excellent low-jitter spectrum can be seen in my review of the Mark Levinson No.30.6 in November 1999 (p.178): the overall noise floor is close to 130dBFS, the central 11.025kHz peak is very narrow, data-related sidebands at 229Hz and its harmonics are all below 120dBFS, and the total jitter is just 153 picoseconds peakpeak.
By contrast, fig.10 shows the jitter spectrum for the Apogee PSX-100's D/A section. The source was the analytical signal stored on a low-timebase-error CD-R, played on a PS Audio Lambda transport connected to the Apogee's S/PDIF data input by 6' of Apature cable. (Because the National Instruments card accepts only single-ended signals, I used the XLR's pin 2 output.) The jitter level is a very high 9.5 nanoseconds (9500ps), and there is a significant rise in the noise floor either side of the tone, implying the presence of low-frequency random jitter, While data-related jitter is fairly lowthe highest-level component at ±229Hz, marked in this graph with a red "11," contributes 370psthere are very strong sidebands present at the harmonically related frequencies of ±164Hz, ±328Hz, and ±492Hz (purple "8," "16," and "22" markers) that contribute almost all of the measured jitter. A pair of AC-power-supplyrelated sidebands at ±120Hz (dark blue "6") can also be just made out.
Fig.10 Apogee PSX-100 D/A, high-resolution jitter spectrum of analog output signal (11.025kHz at 6dBFS with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Source: 16-bit data, PS Lambda transport connected with 6' Apature S/PDIF cable.
To say I was puzzled by this is an understatement. Apogee has an excellent reputation in the pro audio field, and in fact made its name in the 1980s by upgrading Sony PCM-F1 A/D converters with better filters and clocks. (Stereophile's J. Gordon Holt was a big fan of these modifications.) Thinking, therefore, that perhaps the PS transport and the Apature S/PDIF cable might be incompatible with the Apogee, I repeated the measurement using a Meridian 500 transport connected to the processor with 6' of 110 ohm Canare AES/EBU cable. Sadly, the jitter level rose to just under 12.67ns. The ±120Hz components disappeared, and whilemost unusuallythe sidebands at ±164Hz and its harmonics vanished as well, these were replaced with high-level sidebands at ±350Hz, ±700Hz, and ±1051Hz.
It appears that the Apogee's D/A section is unusually sensitive to the data sources and cables with which it is used. As the unit offers a direct link to its D/A section from the A/D section, I was able to perform a third measurement. While I could not synthesize the 229Hz LSB toggling in the analog domain, I used the Audio Precision System One to generate a low-distortion 11.025kHz tone and fed this to the PSX-100's A/D section, sampling at 44.1kHz with 24-bit precision. These 24-bit data were fed straight to the D/A, with the analog output fed, as before, to the Miller Analyzer.
The result is shown in fig.11, with the Meridian/Canare spectrum grayed out for reference. Obviously, there are no data-related sidebands to be seen, but all the high-level sidebands have also disappeared! However, the rises in the noise floor on either side of the tone remain visible. In addition, some power-supply-related sidebands can be seen, at ±120Hz (dark blue "3"), 180Hz (brown "5"), 240Hz (dark blue "6"), and 300Hz (brown "8"). This suggests that, low-frequency noise-related jitter aside, the PSX-100 has problems with its AES/EBU and S/PDIF data receiver circuitry.
Fig.11 Apogee PSX-100, high-resolution jitter spectrum of analog output signal (11.025kHz at 6dBFS). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Source: 24-bit data from Apogee ADC. (Grayed-out trace is 16-bit data from Meridian 500 transport connected with 6' Canare AES/EBU cable.)
Fig.12 shows the frequency response of the A/D section at both 48kHz and 96kHz sampling rates, measured with the Apogee's AES/EBU data output feeding the System One's digital input. By comparison with fig.2, it can be seen that the A/D section is actually flatter up to the band limit than the D/A section. Fig.13 shows the A/D's digital-output amplitude error plotted against analog input level, in dBFS. The error remains below 2dB down to around 115dBFS, which is excellent performance for an ADC.
Fig.12 Apogee PSX-100, A/D frequency response at 1dBFS at 48kHz and 96kHz sampling frequencies (right channel dashed, 2dB/vertical div.).
Fig.13 Apogee PSX-100, A/D departure from linearity, 24-bit data (right channel dashed, 2dB/vertical div.).
Finally, fig.14 shows the action of the UV22 processing on the PSX-100's A/D conversion. The source was an analog 1kHz tone just below the level needed to drive the A/D to full scale (0.2dBFS). The Apogee's AES/EBU output was fed to the digital input of the PrismSound DScope, which was used to average thirty-two 8192-point FFTs on the digital data. (The FFT window used was Prism's "7-term," which has sufficient dynamic range to analyze 24-bit data.) The bottom trace was taken with the Apogee outputting 24-bit data: the noise floor hovers around the 140dBFS level with some very-low-level harmonics visible, the highest of which (the second at 2kHz) almost reaches a still-negligible 100dB (0.001%).
Fig.14 Apogee PSX-100 A/D, digital-domain spectrum of 1kHz sinewave, DC22kHz, at 0.2dBFS (from bottom to top above 15kHz): 24-bit word length; 20-bit word length with UV22 processing; 16-bit word length with UV22 processing (linear frequency scale).
Reducing the output word length to 20 bits but switching in the UV22 processing, which adds narrow-band dither noise at around the Nyquist Frequency, gave the middle trace above 15kHz. The noise floor remains unaffected by the reduction in word length, the UV22 dither doing its stuff. Only when the word length was reduced to 16 bits (top trace) did the noise floor below 15kHz rise, by about two bits' worth. UV22 works, but does not appear to be as effective in preserving audioband resolution as the more aggressive noiseshaped reditherers, such as Meridian's Type D.John Atkinson