Robert Harley posted this a few days ago, and it deserves to be in this section too:
The current issue of TAS (June/July, Issue 182) includes a letter from reader David Sanford questioning whether slight timing errors in digital-to-analog conversion (jitter) is really audible. The letter was sent in reponse to my review of the Esoteric G-0Rb rubidium clock that improves the timing accuracy of D/A conversion. Mr. Sanford found it hard to believe that a shifting of the sample timing (jitter) by several picoseconds would be audible. In my reply, I mentioned that Keith Johnson once told me he could hear the difference between 8 and 15 picoseconds of jitter. (Keith Johnson is the co-inventor of HDCD, recording engineer of all the Reference Recordings projects, designer for Spectral, and has perhaps the most insightful audio mind of anyone working today.)
Keith read the letter and my reply, and sent me this analysis:
David referred to perception of arrival times of sound and to distortion from jitter in a digital system. These are different things not best compared on a pico-second time scale My comments were about perceptual consequences when timing errors are introduced to digital conversion. Sigma delta methods, particularly one bit, introduce much more destructive artifacts. We hear the consequence of jitter rather than jitter by itself.
The time scale argument is interesting:
We can move forward one loudspeaker of a stereo pair 6 inches – an action creating roughly one half millisecond difference in arrival times to create audible consequence.
When delay or arrival times modulate as fast audio range jitter, one finds reasonable perceptual limits require at least one hundred fold smaller moving displacements. Early magnetic recorders demonstrate this disturbance as their unsupported tape paths were long and produced high-speed irregular motion that imparted a sliding paper character to string instrument sound. Manufacturers introduced idler wheels and loop drives to remove sub micro-inches of this scrape flutter and microseconds of modulation noise. We hear the improved clarity along with reduced coloration in the historic development of phonograph records.
At a million times smaller, pico-second changes to sound paths would pose nano-inch listener and speaker constraints – clearly well beyond reasonable contemplation of audibility.
We tend to think of digital processing as on – off and nothing in between. However, modern sigma delta DACs are configured with analog comparators and filters that operate with an environment of continuously changing voltages or currents. When jitter is introduced to clocks or the internal timing of the DAC is not good, the outputs and decision states from these analog functions become cast in stone before or after the time when responses would have been correct. Now a tally or sampling command produces an artifact or error. The jitter not only propagates as itself to cause some distortion but also produces conversion error and both are memorialized in filters involved with ongoing samples. This compound degradation increases at very high sampling rates a factor where larger performance specification numbers may not always provide better performance. We observe spurious signals, cross modulation, pulse response widening and signal coherent noise modulation – things that have unpleasant names and unpleasant sonic consequence.
I think the example of “audible 8 to 15 pico-second jitter” comes from describing a transport evaluation platform at Spectral. Like many digital playback devices, the fixture has multi-bit sigma delta DACs that require very low jitter to work properly. Hence, we provide direct crystal clocking and other technological supports that one should find in any serious product. We can degrade a synchronization and re-clock circuit intended to isolate optical tracking and servo activity of a CD transport. Then, an RF analyzer shows presence of new jitter related to CD condition and oscilloscope traces reveal moving peaks and spikes riding on clock waveforms. An audio analyzer will display upper band noises as well as widened base lines for continuous sine wave spectra. These artifacts might appear inconsequential except that we must consider the averaging nature of analyzer measurements. We have unsteady nasty waveforms that require experimentation and interpretation to reveal the full damage from the tracking interference. A slight increase of noise is observed and careful manipulation of test signals reveals rapidly changing spurious error in the parts per million – the possibly perceptual category. Similar examination methods reveal peak jitter to be well over 5 times the RMS or specification sheet type values produced from the analyzer. Consequences are audible though more so to others at Spectral than to me.
Keith O. Johnson