As development of science and technology, the demand for high-precision time and frequency standard is increasing. The frequency standards of crystal oscillator or rubidium are commonly used as frequency source, but their long-term stability is generally unsatisfactory due to aging and environmental causes. In order to improve time accuracy, we need to evaluate time error exactly. Time error includes system deviation and random deviation. The system deviation includes initial time error, frequency deviation and aging. The random deviation mainly includes five types of power-law noise. The crystal oscillator noise coefficient is estimated and filter estimation of time error is researched in this paper.Firstly the power-law spectrum model of oscillator noise is built in frequency domain. Based on this model, five kinds of noises are analogized. Then the stability of frequency-domain and time-domain is analyzed. The algorithm using transfer function to estimate Allan variance dynamically is put forward and compared with the traditional calculation methods of Allan variance. According to Allan variance curve of time error data, we can use Piecewise-linear fitting method to estimate the type and noise coefficient of oscillator noise.We use least square curve fitting and Kalman filter algorithm to estimate parameter of synchronization time error. Based on Kalman filter algorithm, a new improved adaptive Kalman filer algorithm is put forward. Using this method, the time error, frequency deviation and aging is simulated. Mean while, the Allan variance after controlling is compared with Allan variance before controlling. The simulation result shows that after filter controlling, the synchronization error of oscillator keeps at nanosecond level and the short time stability of oscillator is improved greatly. |