Stochastic Modeling of Short and Long Term Clock Skew

Traditional black box clock skew models are either the power-law noise model or a more recent approach based on auto-regressive (AR) filters. Unfortunately, neither algorithm can accurately model short and long term skew due to limited degrees of freedom or stability constraints. We propose a novel model that employs the current AR algorithm recursively with appropriate pre- and post-processing to achieve numeric stability and accurate reproduction of short and long term effects. The model coefficients are derived from a measured skew signal, with the model output matching an exemplary original signal closely in terms of Allan variance and time-domain behavior.