Every digital audio system has a clock. At 44.1 kHz, it fires 44,100 times per second, once per sample. The assumption is that these pulses are perfectly evenly spaced. They are not.
The deviation from ideal timing is called jitter. It is measured in picoseconds (10^-12 seconds) to nanoseconds (10^-9 seconds), and its effect on the audio signal is not what most descriptions suggest. Jitter does not simply raise the noise floor. It behaves like frequency modulation: every tone in your audio gets surrounded by sidebands at a frequency offset equal to the jitter rate.
Why Jitter Is FM
The connection follows directly from the math of sampling.
A DAC converts stored digital values back to an analog voltage. The timing of each conversion is set by the sample clock. If the clock fires at t = k/f_s + J(t), where J(t) is a small timing deviation, the output sample is read at the wrong moment. From a signal-processing perspective, shifting when a signal is sampled is the definition of phase modulation.
For a sinusoidal tone at frequency f, sampled by a clock with jitter J_peak sin(2π f_j t), the output contains sidebands at f ± f_j with amplitude relative to the carrier:
A_sideband / A_carrier = π × f × J_peakIn decibels relative to the carrier:
L_sideband = 20 × log10(π × f × J_peak)where f is the audio frequency in Hz and J_peak is the peak jitter in seconds.
Take 7 ns peak-to-peak jitter (3.5 ns peak) at a 10 kHz sine wave:
π × 10,000 × 3.5 × 10^-9 = 1.1 × 10^-4 → -79 dBAt 20 kHz with the same jitter:
π × 20,000 × 3.5 × 10^-9 = 2.2 × 10^-4 → -73 dBSideband level scales linearly with audio frequency. Doubling the frequency doubles the amplitude. Jitter always matters most at the top of the spectrum, for exactly the same reason FM synthesis produces louder sidebands when you raise the carrier frequency. The underlying math is identical.
The Audibility Numbers
The most cited empirical study is by Eric Benjamin and Benjamin Gannon of Dolby Laboratories, presented at the 105th AES Convention in 1998 (Preprint 4826). Their blind listening tests established:
- With pure tone test signals: the lowest audible jitter level was 10 ns RMS
- With music: no listener found jitter audible below 20 ns RMS
Ten nanoseconds RMS with a pure tone is already a demanding condition. With real music, psychoacoustic masking raises the threshold further. Program material is not a single tone; the sidebands from jitter land in regions where other audio energy is already present.
Modern asynchronous USB audio interfaces typically measure in the range of tens to a few hundred picoseconds RMS. At 100 ps RMS, you are 100-200× below the Benjamin-Gannon threshold for pure tones. On music, the margin is larger still.
Where Jitter Enters the Chain
Oscillator phase noise. Every crystal oscillator has intrinsic phase noise: random fluctuations in the timing of its output edges. This sets the floor of achievable jitter and is determined by oscillator quality, power supply cleanliness, and mechanical isolation.
S/PDIF clock extraction. S/PDIF embeds the sample clock in the data stream. The receiving device must extract it using a phase-locked loop (PLL). PLL jitter depends on loop bandwidth and input signal quality. Long cable runs, ground loops, and marginal connectors increase extraction jitter. S/PDIF carries no separate clock line, so the recovered clock is only as clean as the PLL allows.
USB transport. Standard synchronous USB audio depends on host timing, which carries jitter from the OS scheduler and USB controller. Asynchronous USB reverses the relationship: the audio interface controls its own clock and tells the host how fast to deliver data, decoupling the audio sample clock from USB timing entirely. Essentially all modern audio interfaces use asynchronous mode.
Daisy-chaining. Word clock signals accumulate jitter as they pass through each device. A single output driving five inputs through a chain of cables and PLLs compounds jitter at every hop. A dedicated clock distribution box with buffered individual outputs avoids this.
Random Versus Correlated Jitter
The spectrum of the jitter matters as much as its amplitude.
Random jitter has no relationship to the audio content or digital bit patterns. Its sidebands are incoherent: they average out into a broadband noise contribution rather than discrete spectral lines. The effect on a measured noise floor is real but small.
Correlated jitter has a deterministic component that repeats with data transitions or audio content. Because it is periodic, it produces coherent sidebands at specific frequency offsets that do not average down. Correlated jitter at the same RMS amplitude as random jitter is more audible precisely because its artifacts sit at fixed spectral positions, where they become perceptible when they align with harmonics or other tonal content.
S/PDIF clock extraction under marginal conditions, or PLLs with insufficient loop bandwidth, tend to produce correlated jitter because bit transition patterns partially determine when the PLL adjusts its phase.
Modern Specs in Context
RME's SteadyClock FS, a second-generation reclocking circuit used across their current interface range, claims sub-picosecond jitter in synchronous operation. That is more than three orders of magnitude below the Benjamin-Gannon audibility threshold.
The practical conclusion: on any modern interface using asynchronous USB or a dedicated internal clock, jitter is not the limiting factor in your signal chain. Room acoustics, monitoring accuracy, and microphone placement all contribute more to what you actually hear.
Clocking becomes relevant in specific scenarios: multi-device setups using S/PDIF or ADAT without asynchronous sample-rate conversion, older equipment with lower-quality oscillators, or any situation involving long chains of digital connections where jitter accumulates without reclock stages.
The Audiophile Problem
The high-end audio market has produced strong claims that expensive external word clock generators audibly improve modern interfaces. Some of these claims rest on subjective impressions gathered under sighted conditions.
The measurable reality is harder to reconcile with those claims. Jitter measurements of current converters run one to three orders of magnitude below published audibility thresholds. The audible differences between a $200 interface and a $2000 interface are often real; whether clock jitter is the responsible mechanism is a separate question that measurements consistently answer in the negative.
Understanding the FM connection explains both why jitter theoretically matters, because it creates genuine spectral artifacts, and why it practically does not at modern performance levels: those artifacts sit 40-60 dB below the masking threshold on any program material with normal spectral content.