Wadia 121decoding Computer D/A processor Measurements
Sidebar 3: Measurements
I examined the Wadia 121's electrical performance with Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (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. To test the USB input, I played AIFF and WAV files with iTunes and Bias Peak Pro running on my MacBook Pro, using the AudioMIDI Set-up utility to ensure that the sample rate and bit depth were set correctly. Apple's USB Prober utility identified the 121 as "Wadia USB Audio 2.0" and confirmed that it operated in isochronous asynchronous mode with a 24-bit word length. The Wadia's USB input operated correctly at sample rates up to 192kHz, including 88.2kHz and 176.4kHz, as did the S/PDIF and AES/EBU inputs.
The output level from the RCA jacks can be set to 1.0V, 2.0V, or 4.0V, using a sequence of button pushes on the remote control. (The default appears to be 2V.) The maximum level at 1kHz, with the volume control set to its maximum, was the same from the balanced XLRs and single-ended RCAs. It was 3.48V with the output set to "4.0V," 1.74V with it set to "2.0V," and 871mV with it set to "1.0V." Each of these maximum outputs is 1.2dB lower than the specified level. The output impedance was a low 48 ohms from the RCA jacks, and twice that value from the XLRs, as expected. Both pairs of outputs preserved absolute polarity (ie, were non-inverting) with the front-panel Phase LED off. The XLRs are wired with pin 2 hot. The volume control operated in accurate 0.5dB steps.
Fig.1 shows the Wadia 121's impulse response with 44.1kHz data (one sample at 0dBFS). The processor uses Wadia's proprietary DigiMaster reconstruction filter, which made its debut in the company's groundbreaking Model 2000 Decoding Computer, from the late 1980s. Fig.1 confirms that this filter offers superbly well-defined time-domain performance, with just one cycle of ringing before and after the impulse. However, the trade-off for this performance is less effective suppression of ultrasonic images in the frequency domain.
This is shown graphically in fig.2, a test pioneered by MBL's chief engineer Jürgen Reis. The red trace is a wideband spectrum of the Wadia's output while it decoded a full-scale 19.1kHz tone sampled at 44.1kHz. The aliasing product at 25kHz (44.119.1kHz) is suppressed by just 10dB. While the distortion harmonics at 38.1 and 57.3kHz are very low in level, some higher-frequency aliasing products are also visible. The blue trace in fig.2 is the spectrum of white noise, again sampled at 44.1kHz. This trace reveals the slow ultrasonic rolloff of the DigiMaster filter, and shows that while there is a null at 44.1kHz, the suppression is less good in the octave above that frequency.
The filter's slow rolloff means that with 44.1kHz data (fig.3, green and gray traces), the Wadia 121's frequency response is down by 1dB at 15kHz, and by 3dB at 20kHz. This slight lack of top-octave energy won't be audible. However, what surprised me about this graph was that the response with data sampled at 96kHz (cyan and magenta traces) was almost identical to that at the lower sample rate. Only with 192kHz data (blue and red traces) did the Wadia's response extend above the audioband, but with then a very sharp cutoff just below half the sample rate. (Wadia subsequently confirmed that this behavior at the three sample rates is correct.) Channel separation (not shown) was excellent at >115dB below 1kHz in the RL direction, but about 7dB less good in the other direction.
I got some anomalous results when I examined the Wadia's resolution. Fig.4 shows the spectrum of the 121's output while it decoded dithered 16-bit data (cyan and magenta traces) and 24-bit data (blue and red traces) representing a 1kHz tone at 90dBFS. The increase in bit depth drops the noise floor by around 9dB, implying ultimate resolution between 17 and 18 bits. To generate this graph, I fed the data to the Wadia from the Audio Precision using an AES/EBU link. To my astonishment, when I repeated the analysis using a coaxial S/PDIF link to transmit the 24-bit data, I got 16-bit resolution. The blue and red traces in fig.5 repeat the spectrum with 24-bit data and an AES/EBU link; the cyan and magenta traces in this graph were taken with the 24-bit data transmitted with the coaxial S/PDIF link. I repeated the analysis using a TosLink connection from the Audio Precision, but with no difference in the result. To check that the Audio Precision was working properly, I then used a TosLink connection from my MacBook Pro. However, I got the same result: 24-bit data but 16-bit resolution. Finally, I used a USB connection from the laptop, and although I made sure that the connection was correctly set to transmit 24-bit integer data, the noise floor was around the 15-bit level (not shown).
This lack of resolution can be seen in fig.6, which shows the waveform of an undithered sinewave at exactly 90.31dBFS. The three DC voltage levels described by the data are obscured by high-frequency noise.
Moving on to distortion, the Wadia 121 offered very linear performance (fig.7). A full-scale 50Hz tone was accompanied by just the second harmonic at a very low 110dB (0.0003%) in both channels, and the third harmonic at a slightly higher level in the right channel only. This graph was taken into a high 100k ohm impedance; reducing the load to the punishing 600 ohms increased the level of the third harmonic to a still-low 103dB (0.0007%) in the right channel, 110dB in the left. Intermodulation distortion was similarly very low in level (fig.8), but you can see that the "leaky" reconstruction filter allows the images of the 19 and 20kHz tones to be reproduced at almost full level.
With 16-bit data and an AES/EBU connection (fig.9, cyan and magenta traces), while the levels of the odd-order harmonics of the LSB-level squarewave are not accentuated, these harmonics are obscured by a rise in the noise floor around the 11.025kHz tone similar to that seen in the intermodulation graph, and a pair of sidebands appears at ±60Hz. With 24-bit data (blue and red traces), the odd-order harmonics have disappeared but the noise floor remains at the 16-bit level.
That the AES/EBU input behaved correctly but the other inputs didn't is extremely unusual. I suspect that there was some kind of problem with our review sample. Jon Iverson used the S/PDIF and USB connections to drive the Wadia 121 rather than AES/EBU; I do wonder, therefore, whether his feeling that the processor had a slightly veiled quality was related to this.
Although JI was very impressed by the Wadia 121's sound with headphones, I didn't measure the performance from its headphone output, as I was increasingly becoming convinced that there was something wrong with our review sample when it was on my test bench. I will report on that aspect of the Wadia's performance in a Follow-Up review when we receive a second unit.John Atkinson