Abstract: High-accuracy auto-time synchronization is the foundation of the navigation constellation and a key technology of auto-navigation.The accuracy of time synchronization among navigation satellites is directly affected by frequency stability of on-board atomic clocks.GPS Rubidium atomic frequency standards(RAFS) exhibit non-trivial aging and aging noise characteristics,whereas the Allan variance does not explicitly converge for the noise type of α not more than-3),and can be greatly affected by frequency drift.The three-sample variance,commonly referred as a renormalized Hadamard variance was proposed,which was unaffected by linear frequency drift,converges for α more than-5),and thus can be used for modeling noise in Rubidium frequency standards.A standard Kalman filter algorithm based on improved clock error predication model and inter-satellites measurements equations was established for practical use in on-board RAFS.System simulation results show that more precise auto-time synchronization is achieved by this method.