Modified Allan variance

The Allan variance has a drawback in that it is unable to separate the white phase modulation (WPM) from the flicker phase modulation (FPM). Looking at their response to Power-law noise it is clearly seen that WPM and FPM have almost the same response to tau, but WPM is linearly sensitive to the system bandwidth f_H whereas FPM is only weakly dependent on it. Thus, by varying the system bandwidth the WPM and FPM noise forms may be separated. However, it is impractical to alter the hardware of the measurement system. By post-processing the sample-series and implementing a software bandwidth a modified Allan variance measure can be given capable of resolving the noise forms.

The modified Allan variance is defined for [3] using time error samples as

${\displaystyle \operatorname {mod} \sigma _{y}^{2}(n\tau _{0})={\frac {1}{2\tau ^{2}}}\left\langle \left[{\frac {1}{n}}\sum _{i=0}^{n-1}x_{i+2n}-2x_{i+n}+x_{i}\right]^{2}\right\rangle }$

or with average fractional frequency time series and τ = nτ₀

${\displaystyle \operatorname {mod} \sigma _{y}^{2}(n\tau _{0})={\frac {1}{2}}\left\langle \left[{\frac {1}{n}}\sum _{i=0}^{n-1}{\bar {y}}_{i+n}-{\bar {y}}_{i}\right]^{2}\right\rangle ,}$

where n is the integer number of samples averaged over.

The modified Allan variance estimator for time error time series is

${\displaystyle \operatorname {mod} \sigma _{y}^{2}(n\tau _{0})={\frac {1}{2n^{4}\tau _{0}^{2}(N-3n+1)}}\sum _{j=0}^{N-3n}\left\{\sum _{i=j}^{j+n-1}x_{i+2n}-2x_{i+n}+x_{i}\right\}^{2}}$

or with average fractional frequency time series

${\displaystyle \operatorname {mod} \sigma _{y}^{2}(n\tau _{0})={\frac {1}{2n^{4}(M-3n+2)}}\sum _{j=0}^{M-3n+1}\left\{\sum _{i=j}^{j+n-1}\left(\sum _{k=i}^{i+n-1}{\bar {y}}_{k+n}-{\bar {y}}_{k}\right)\right\}^{2}}$