Wadia 850 CD player Measurements part 2
The 850 offers good linearity, any level error (fig.5) lying between -1dB and +1dB down to -110dBFS. The waveform of an undithered 1kHz sinewave at -90.31dBFS (fig.6) has a reasonably well-defined shape, though there is more noise than with the ultraquiet Meridian 508.24 (see WP's review elsewhere in this issue). Increasing the word length to 20 bits gave a good sinewave shape (not shown), though this was still overlaid with some audioband noise, again implying that the 850's ultimate resolution is limited by analog or dither noise rather than by the intrinsic performance of the digital filter or the digital volume control.
Fig.5 Wadia 850, departure from linearity (2dB/vertical div., right channel dashed).
Fig.6 Wadia 850, waveform of undithered 1kHz sinewave at -90.31dBFS (16-bit data).
The "leaky" nature of the Wadia's DigiMaster anti-imaging filter can be seen in fig.7, which shows the spectrum of the player's output while it decodes 16-bit data representing a full-scale mix of 19kHz and 20kHz tones. (Each individual tone lies at -6dBFS; the combined waveform just reaches 0dBFS.) While the conventional first- and second-order intermodulation products at 1kHz, 18kHz, and 21kHz are all well down in level, implying bombproof analog circuitry, a number of aliasing products are visible in this graph, resulting from the 19kHz and 20kHz tones reflecting back from the 44.1kHz sampling frequency and its harmonics. That this is the case can be demonstrated by changing the sampling frequency to 48kHz (fig.8): the intermodulation products remain unchanged in frequency; the aliasing products follow the change in sample rate.
Fig.7 Wadia 850, HF intermodulation spectrum, DC-22kHz, 19+20kHz at 0dBFS, 44.1kHz sampling frequency (linear frequency scale, 20dB/vertical div.).
Fig.8 Wadia 850, HF intermodulation spectrum, DC-22kHz, 19+20kHz at 0dBFS, 48kHz sampling frequency (linear frequency scale, 20dB/vertical div.).
The subjective effect of this filter behavior is hard to predict. Certainly these aliasing products will not be audible in themselves. But look at fig.7—metal-dome tweeters with their "oil-can" resonances in the 22-25kHz range are not going to like being hit with such high levels of energy in this region. I do wonder if the music-dependent brightness that I occasionally noticed with the Wadia was due to this sort of behavior.
I used the Miller Audio Research analyzer to look at the Wadia's jitter performance (footnote 1). This drives the CD player under test with data representing a high-level 11.025kHz (Fs/4) tone and a 229Hz tone toggling the LSBs on and off, while it performs a high-resolution FFT analysis—32,768 points, with 64 FFTs averaged—to examine the noise floor in the analog domain.
The Wadia's absolute timebase error was a negligible +5 parts per million! The jitter was also low at a calculated 167.6 picoseconds peak-peak. Fig.9 shows the FFT-derived spectrum of the Wadia's analog output, centered around the test signal's 11kHz center frequency. For comparison, the grayed-out trace shows the spectrum of the Meridian 508.24's output spectrum taken under identical conditions. You can see that the Wadia is 2-3dB noiser than the Meridian below the central frequency, and about 1dB noiser above it. There are a number of discrete noise components present, indicated with light blue numbers, but these are very low in absolute level. The spectral lines marked with red numbers are related to the 229Hz signal frequency and contribute most of the jitter content. However, there are some low-frequency components present at ±15Hz, ±30Hz, ±171Hz, and ±189Hz, marked with purple numbers "1," "3," and "5." I have no idea what these are due to, nor would I like to conjecture about their subjective consequences. You can see, however, that the Meridian's output contains higher levels of very-low-frequency jitter.
Fig.9 Wadia 850, high-resolution jitter spectrum of analog output signal (11kHz at -10dBFS with LSB toggled at 229Hz). Center frequency of trace, 11kHz; frequency range, ±3.5kHz. Grayed-out spectrum is that of the Meridian 508.24.
Finally, I tested the Wadia's error correction using the Pierre Verany test CD, which has deliberate gaps in the data spiral. The player would track the disc without audible dropouts through track 35, which has a gap in the data 2.4mm long. This excellent performance is among the best I have measured.—John Atkinson
Footnote 1: Stereophile used to examine digital wordclock jitter at the DAC, using a Meitner LIM analyzer, and while this produced consistent results, it is the effect of jitter in the analog domain that really matters. We have therefore switched to using the Miller analyzer.—John Atkinson