Assessing the Validity of Suboptimal 2-State Clock Models

Abstract : In real-time Kalman filter applications involving clocks, it is common practice to model the clock errors with simply two state variables, namely phase and frequency. It has been pointed out in previous PTTI papers that such models are suboptimal because they cannot accommodate flicker noise components exactly, Thus it is desirable to have a means of assessing the "goodness of fit" of a proposed 2-state model in the particular application at hand. The methodology suggested here involves first choosing a higher-order model which is used as a truth model for purpose of analysis. The suboptimal gains generated by the 2-state model being assessed are then cycled through the truth model to generate realistic error covariances to be associated with the suboptimal filter. This method of assessing the degree of suboptimality of the 2-state model is especially useful in prediction application where the At interval is relatively large. Subtleties and possible pitfalls in the design of the truth model are discussed in the paper, and examples are presented.