In this paper we propose an approach for performing fault detection and identification in clock ensembles based on the generalized likelihood ratio test. We show that by applying a set of purposefully-designed statistical tests, one can successfully detect faults occurring in a clock of the ensemble, and identify which measurement in the ensemble is most likely to have triggered the detection. We first develop the theoretical framework for the characterization of the detectors and their performance, and validate the derivations via Monte Carlo simulations. Then, we apply the statistical tests to an ensemble of cesium clocks, aiming at detecting and identifying three types of non-nominal behaviors. The faulty conditions are obtained by injecting a pattern of phase steps, a phase and frequency drift, and an oscillatory phase component.
Read full abstract