What Is Allan Deviation (ADEV)?

By KCCS Engineering Team · Updated 2026 · 6 min read

Allan deviation (ADEV, σ(τ)) is the standard way oscillator manufacturers specify short- and medium-term frequency stability. If you have read an OCXO datasheet, you have seen numbers like "1.03E-13 @ 1 s" or "8E-14 @ 10 s". This note explains what those numbers actually mean, how to read an ADEV plot, and what to look for when comparing oscillators.

What Is Allan Deviation?

Allan deviation quantifies how much an oscillator's fractional frequency changes when averaged over a measurement interval τ. It is computed by measuring frequency in consecutive, non-overlapping blocks of duration τ, then taking the standard deviation of adjacent differences.

Why use adjacent differences instead of plain standard deviation? Because oscillators have drift (aging) that would otherwise dominate the number. The Allan difference cancels out linear frequency drift, revealing the noise floor that actually matters to timing applications.

How to Read the Number

ADEV is expressed as a dimensionless fractional frequency. A value of 1E-12 at τ = 1 s means that, after averaging the frequency for one second, it differs from the mean by about 1 part in 10¹². On a 10 MHz oscillator, that is 0.01 Hz.

Why τ Matters — Don't Compare Apples to Oranges

Oscillator noise behaves differently at different averaging times. At very short τ (milliseconds), white phase noise dominates and ADEV decreases as τ increases. In the mid-range (1 ms to 10 s), flicker phase noise flattens the curve. At long τ (100 s and beyond), random walk and aging take over and ADEV starts rising again.

This is why two datasheets quoting "1E-11" do not necessarily compare directly — one might be at τ = 1 s and the other at τ = 100 s. Always compare at the same averaging time, or better, look at the full ADEV plot.

How ADEV Connects to Your Application

PTP/IEEE 1588 boundary clocks care about ADEV at τ = 1 s and τ = 10 s, because the servo updates on that timescale.

Holdover after GNSS loss depends on temperature stability and aging, not ADEV alone. ADEV tells you short-term noise; holdover tells you long-term drift.

Frequency counters and calibration standards need the lowest possible ADEV across the full τ range, because the measurement gate length varies.

ADEV vs Phase Noise — Two Sides of the Same Coin

Phase noise (dBc/Hz at offset f) describes the spectral purity of the carrier. ADEV describes time-domain stability. They are mathematically related but not interchangeable. A low phase noise at 10 kHz offset does not guarantee low ADEV at τ = 1 s. If your application is sensitive to timing jitter, look at phase noise. If it is sensitive to frequency drift over seconds, look at ADEV.