jitter is simply undesired variations in clock frequency that can make it hard to get truly excellent sound from professional and consumer digital audio systems. Although the purity and stability of clocks are critical for the quality of digital playback, most USB-S/PDIF interface manufacturers do not publish any jitter numbers at all.
An excellent review of several popular USB-S/PDIF
transports in the December, 2010 Stereophile contains part of the answer why this
is the case: Most products tested showed marginal jitter performance, which was
directly correlated with audible deficiencies in sound quality.
Making accurate jitter measurements—typically handfuls of picoseconds—is much more difficult than the usual frequency response specs.
Highly-specialized test gear, and expertise in interpreting the data, are both
required to obtain reliable results. Audiophilleo has developed special testing
protocols using the WaveCrest DTS timing analyzer and the TSC 5120A phase
noise analyzer.
A complete suite of such laboratory-grade instruments can cost over
$100,000, but
to get accurate results, it's what's required to measure both period jitter as
well as integrated phase noise (phase jitter) values over a 1 Hz to 100 kHz bandwidth.
The bandwidth over which the jitter is measured is an important parameter, and
allows for meaningful comparison between products. Also keep in mind that some
jitter specs are taken from internal test points: The Audiophilleo numbers are
measured on the actual BNC output connector, which gives a more realistic
picture of likely real-world performance.
Audiophilleo1 and Audiophilleo2 jitter specifications
Here's a typical phase noise plot for the Audiophilleo1/2. It shows just 3.8 ps RMS phase noise integrated from 1 Hz to 100 kHz taken with a Symmetricom 5120A from a standard production (serial 0029) Audiophilleo1.
Note that two types of measurement are relevant: The simple period jitter only
tells part of the story; integrated phase jitter with bandwidth specifications
is highly relevant, because low frequency or long-term jitter is also
potentially harmful. Most DACs are typically less immune to this form of clock
variation.
Commonly-used, general-purpose test devices such as the Prism dScope and Audio
Precision SYS 2722 are not ideal for characterizing state-of-the-art USB-S/PDIF
converters, because their timing accuracy is, respectively, 2000 ps (2
nanoseconds) and 1000 ps (1 nanosecond) between 50Hz and 100Khz. This limits
their usefulness for measuring the ultra-low jitter inherent in the Audiophilleo
design.
Testing methodology
Here’s how we performed the testing. The Symmetricom TSC 5120A phase noise
analyzer—specialized for clock measurement—gathers a large number of data
points, which are then integrated in software to produce a performance graph
across the desired bandwidth, 1 Hz to 100 kHz in this case.
Because the TSC 5120A measures clock waveforms, rather than the more
complicated, data-with-embedded-clock waveforms of
S/PDIF, we used the
Audiophilleo’s internal software to generate periodic clock signals, which it
can do just as easily and accurately as it does BMC-encoded. Exactly the
same electronics and signal pathways are used in either case.
In this way, we measured various carrier frequencies, and calculated phase
jitter measurements from 10 Hz to 100 kHz and 1 Hz to 100 kHz. The numbers
published here reflect worst-case measurements. Although this test includes noise induced by the power supply, output stage, and crystal clocks,
it does not measure data correlated jitter. We believe that this is not an issue because our design should have negligible data induced jitter. Tests
show that high frequency jitter only increases less than 1ps with various data patterns.
Why does jitter matter?
In practical terms, jitter refers to a variety of unwanted variations in control
signals—the clocks—that tell digital audio systems what to do and when. There
are several causes of clock jitter, but regardless of the origin, when clocks
start going slightly awry—even handfuls of trillionths of a second—sound quality
may ultimately be degraded. For a comprehensive look at the pernicious effects
of jitter, see "The Digital Enemy" by TNT Audio.
Only the grossest jitter would be capable of harming the actual transmission of
digital audio data. The AES/EBU specification for S/PDIF allow for up to 80 nS
of jitter before bits are lost.
So if all the bits are getting through intact, why is jitter a problem, and why
does it need to be so low for the best sound?
jitter levels capable of causing data corruption are typically 100 times greater
than those which cause problems for audio enthusiasts. jitter that creates
distortion in the audio spectrum is a more subtle phenomenon, and relates to how
S/PDIFdigital communication works, and its effects upon digital-to-analogue
conversion. Problems arise because S/PDIF combines clocking information and data
into a single digital stream of data that runs down one wire. The DAC has to
extract the clock signals from this incoming bit stream, and uses them to
control the conversion process. A noisy or jittery S/PDIF clock means the
conversions do not occur at precisely the right time, which in turns leads to
disagreeable distortions that show up in the audio frequency range. Furthermore,
many jitter-caused artifacts are not hidden by the natural harmonic structures
of music, and thus tend to stand out.
How much jitter is too much?
There are many opinions, but jitter over,
say, 100 picoseconds is fairly easy to hear on very good systems. The effects
and audibility of jitter vary considerably, depending upon:
See the Articles page for more information.