An adaptive oscillator noise analysis using factor analysis

Abstract

Multivariance oscillator noise analysis methods provide precise measurements of some noise sources of a high-performance oscillator, while others are inaccurate due to poor observability and an ill-conditioned problem. The ill-conditioned problem can be solved by separating the averaging time manually. The disadvantage of this strategy is that difficulties in execution arise when measuring noise sources from dynamic variances. This paper introduce an adaptive noise intensive coefficients estimator which separates the averaging time automatically and thus is executable on dynamic variances. Two cost functions, one of deterministic linear frequency drift and the other of noise intensity coefficients, are introduced to form the estimator. The estimator is tested with simulated data. In addition, a dynamic total variance is computed from more than 8.2 years of GPS clock data, and 31 765 groups of noise intensity coefficients are estimated using the adaptive estimator. The results show that the adaptive noise intensity coefficient estimator is as precise as the multivariance oscillator noise analysis methods and is robust.