Digital Processor Reviews

In the November 1994 issue of Stereophile, Robert Harley reviewed three products that were intended to reduce word-clock jitter in an S/PDIF or AES/EBU serial datastream: the Audio Alchemy DTI Pro, the Digital Domain VSP, and the Sonic Frontiers UltrajitterBug. (All three handle 16-bit data sampled at rates of 44.1kHz and 48kHz.) This review was published before Paul Miller and the late Julian Dunn created the J-Test signal, which combines a high-level tone at exactly one quarter the sample rate (Fs/4) with a low-level squarewave at 1/192 the sample rate produced by toggling the LSB. (This signal is not dithered, as this would interfere with its diagnostic function—the level of the noise floor between the harmonics of the squarewave is due to the processor's analog noise.)

I still have Stereophile's review samples of the Sonic Frontiers UltraJitterBug and Digital Domain VSP, so I performed some measurements using a 16-bit J-Test signal, generated by my Audio Precision SYS2722 and transmitted to the D/A processors via a 15' optical S/PDIF link. (As the UJB doesn't have optical S/PDIF outputs, the outputs of both processors were taken to the DACs via a short coaxial datalink.)

The first question was what D/A processors to use for the tests? In his review Robert Harley had used a PS Audio UltraLink, which had relatively poor jitter rejection. I dug around in my storage unit and found that UltraLink sample, along with the Parts Connection Assemblage DAC-1 kit, which I had assembled back in the 1990s and was reviewed by the late Wes Phillips in April 1995. For completeness sake, I also examined the effect of the UJB and VSP with a more recent product, an inexpensive E-Mu 0404 A/D and D/A converter box.

Looking first at the performance with the Assemblage DAC-1, fig.1 shows a narrowband spectrum of its output when fed 16-bit coaxial J-Test data. The sloping green line shows the correct levels of the harmonics of the low-frequency squarewave-these are all reproduced too high in level, especially those closest to the spike that represents the Fs/4 tone, due to jitter. These were even higher in level via the optical datalink and inserting the Sonic Frontiers UJB had no effect (fig.2).

Replacing the UJB with the Digital Domain VSP with its jitter-reduction button pressed, which processes the incoming data with an asynchronous sample-rate converter chip before sending it to the Assemblage, gave the spectrum shown in fig.3. It looks as if the squarewave harmonics are now at the correct level but are obscured by a noise floor that is around 12dB higher in level than in figs.1 and 2. (This is presumably due to the mathematical limitations of the DSP chip.) In addition, new sideband pairs make an appearance, at ±1375Hz and ±2746Hz. A sideband pair at ±50Hz can also be seen in the left channel (blue trace).

The changes in the behavior of the E-Mu DAC with the two jitter-reduction boxes was similar. With the PS Audio UltraLink, however (fig.4), while the effect of the VSP (fig.5) was the same as what it had been with the Assemblage—a reduction in the levels of the squarewave harmonics, a higher analog noise floor, and new sidebands at ±1375Hz and ±2746Hz—the UltraJitterBug (fig.6) increased the levels of the sidebands closest to the Fs/4 tone. While this is not what I had found with the Assemblage DAC-1 (fig.2), this was not what I was expecting from a device that is supposed to reduce jitter.

For my final series of tests, I used the Audio Precision SYS2722's ability to output a serial datastream corrupted with specific amounts and types of jitter. I created S/PDIF data representing a full-scale, 16-bit 10kHz tone and sent it to the PS Audio Ultralink via the 15' optical link, first without jitter, then afflicted with jitter at 1kHz with an amplitude of 1 nanosecond.

Fig.7 shows the spectrum of the PS Audio's output converting the 10kHz data to analog. The noise floor is even and close to the level of the dither in the 16-bit test signal, though some low-level enharmonic spikes are present. By contrast, with the jittered tone (fig.8), strong sidebands have appeared at ±1kHz and the noisefloor shows significant modulation, acquiring a strange scalloped appearance.

I then repeated the test, transmitting the data first by the Sonic Frontiers UJB (fig.9) then by the via Digital Domain VSP (fig.10). You can see that the UJB has not eliminated the jitter. The sidebands at ±1kHz are as strong as they were in fig.8, though the low-level enharmonic spikes have been removed. The VSP, however, has completely got rid of the 1kHz sidebands, though this is at the expense of a raised noise floor and the generation of sidebands at ±1375Hz.

Conclusion: With the 20/20 hindsight offered by these 2020 tests, it looks as if the Sonic Frontiers UltraJitterBug doesn't so much eliminate jitter in the S/PDIF datastream as replace it with its own jitter signature. The Digital Domain VSP's asynchronous sample-rate converter certainly eliminates incoming jitter, but by raising the noise floor reduces the resolution of the data. And. Of course, neither the UJB or VSP can reduce the damage done to the decoded audio by jitter that is generated within the D/A processor itself.

The real solution to the problem products like these were intended to address a quarter-century ago was for the designers of D/A processors to incorporate data receivers with much better jitter rejection than those used by the legacy models I used for these tests. See, for example, the exemplary manner in which a modern DAC, the dCS Bartók, handles 16-bit J-Test data (fig.11).—John Atkinson