| Global Navigation Satellite System?GNSS?is a system based on time measurement.The on-board atomic clock acts as the time reference on the satellite used for distance measurement of the system,and it is also one of the core payloads of GNSS.Therefore,the performance of on-board atomic clock directly determines the precision of navigation and positioning.Obviously,it has important theoretical significance and practical value to carry out related researches on GNSS satellite clock for normal maintenance and operation of the system.In these related researches,evaluating and analyzing the performance of GNSS on-board atomic clock is an important means to know the operational condition of satellite clocks.Modeling and prediction of GNSS satellite clock bias?SCB?are also important aspects of studying GNSS satellite clock,because it plays a significant role in maintaining time synchronization of the system and meeting the needs of real-time kinematic precise point positioning?PPP?.Meanwhile,the above-mentioned researches are mainly conducted based on SCB data products.As a result,reasonably assessing the quality of SCB products and effectively preprocessing SCB data are the precondition for carrying out the study.In this dissertation,GNSS SCB products are used to systematically study the preprocessing methods used for SCB data,quality evaluation of SCB products,performance evaluation of satellite clocks,and modeling and prediction of SCB.The main achievements and contributions are summarized as follows:1.Based on the median absolute deviation?MAD?method,a new data preprocessing method is proposed for long-term SCB data.Specifically,nonempty data is extracted first from the original long-time SCB sequence to be preprocessed using the MAD method.The subsequent step is to preprocess again the daily data of the preprocessed SCB data based on the MAD method.Using the proposed method to preprocess the one-year BDS SCB data,the results show that the new preprocessing method is effective.2.A preprocessing method of abnormal values in SCB data is proposed by using a wavelet analysis method.Specifically,the frequency data corresponding to original SCB data are decomposed by wavelet analysis at first,obtaining a corresponding low-frequency coefficient and corresponding high-frequency coefficients at all levels from the wavelet decomposition.Then the location of abnormal points in the coefficients is identified and the abnormal points are processed by combining wavelet coefficient diagrams with threshold formulas.Finally,the processed SCB data eliminating abnormal values are obtained by reconstructing the processed wavelet coefficients.The new method and its related characteristics are tested and analyzed from three aspects,including the processing effect of outliers,the effect exerted by different scales of wavelet decomposition on processing results and the different processing effect generated while using different wavelet functions.3.Three-year precise satellite clock data products derived from multi-satellite orbit determination are used to evaluate the performance of BDS satellite clocks.Specifically,the characteristics of SCB data are discussed by using the improved MAD method to preprocess original SCB data.Long-term variations of satellite clocks'phase,frequency,clock drift,and model noise levels are analyzed based on the quadratic polynomial SCB model.Periodicity of BDS SCB is analyzed by using spectral analysis method.Frequency stability of BDS satellite clocks is calculated and discussed based on Overlapping Hadamard Variance.Frequency accuracy and daily frequency drift of the clocks are calculated to analyze their long-term variations.Integrating the aforementioned discussions and corresponding experiment results,the long-term performance of BDS satellite clocks is relatively comprehensively evaluated and analyzed.4.The long-term performance of GPS BLOCK IIF satellite atomic clocks is analyzed and evaluated based on long-term variations of five performance indexes,including frequency accuracy,frequency drift,frequency stability,noise level and clock bias's periodic terms.The analyzed and evaluated results show that frequency accuracy,short-term and daily drift,and average noise level of the rubidium clocks are respectively 7.1?10-12(?2.1?10-13),3.4?10-19(?1.2?10-20)/second and 5.5?10-14(?1.1?10-14)/day,and about 0.2 nanoseconds.The corresponding indexes of the cesium clocks are respectively 1.0?10-12(?2.9?10-15),1.4?10-18(?4.9?10-20)/second and 3.4?10-15(?5.4?10-16)/day,and about 1.0 nanosecond,and the change of these indexes is relatively stable.The frequency stability of rubidium clocks for averaging time of 2 hours,6 hours,12 hours and 24 hours??=7,200 s,21,600 s,43,200 s and86400 s?are 3.4?10-14,2.3?10-14,7.3?10-15 and 6.4?10-15 and the corresponding stability indexes of the cesium clocks are 1.9?10-13,1.1?10-13,7.9?10-14 and 5.6?10-14.In addition,there are obvious periodic terms in satellite clock bias and their primary periods are approximately equal,i.e.one-half,equal or double to the corresponding satellite orbit periods.5.In order to better express characteristics of SCB and improve the SCB prediction precision,a new SCB model is put forward which will consider the physical characteristics of satellite atomic clock,the cyclic variation and the random part of SCB.First,the new model employs a quadratic polynomial model with periodic item to fit and extract the trend term and cyclic terms of SCB.Then based on the characteristics of fitting residuals,a time series ARIMA model is used to model the residuals and the results from the two models are combined to obtain final SCB prediction values.At last,the paper uses precise SCB data from IGS to conduct prediction tests,and the results show that the proposed model is effective and has better prediction performance compared with the quadratic polynomial model,grey model and ARIMA model.Also,the new method can overcome the deficiency of ARIMA model,i.e.the inaccurate model recognition and its order determination.6.A new SCB model is proposed,which can take the physical feature,cyclic variation and stochastic variation behaviors of the on-board atomic clock into consideration by using a robust least square collocation?LSC?method.The proposed model firstly uses a quadratic polynomial model with periodic terms to fit and abstract the trend term and cyclic terms of SCB.Then for the residual stochastic variation part and possible gross errors hidden in SCB data,the model employs a robust LSC method to process them.The covariance function of the LSC is determined by selecting an empirical function and combining SCB prediction tests.Using the final precise IGS SCB products to conduct prediction tests,the results show that the proposed model can get better prediction performance and more accurately express characteristics of SCB.Specifically,the prediction precision of the results can enhance 0.457 nanoseconds and 0.948nanoseconds respectively,and the corresponding prediction stability can improve 0.445nanoseconds and 1.233 nanoseconds,compared with the results of quadratic polynomial model and grey model.In addition,the results also show that the proposed covariance function corresponding to the new model is reasonable.7.An improved strategy for the prediction principle based on single difference values of SCB is advanced,and a preprocessing method specific to the single difference data is designed.Based on SCB prediction tests,this paper analyzes the prediction performance of several commonly-used prediction models under the condition of single difference data of SCB.On that basis,the paper theoretically derives and certifies that the single difference sequence of SCB can be accurately expressed by the liner model,and a prediction method used for IGS RTS SCB correction is given based on the derived and certified results by combining the analysis for the SCB correction.The prediction tests show that the prediction precision of the method can reach0.06 nanoseconds in the prediction length of 30 seconds.8.An improved prediction model is proposed in order to enhance the prediction performance of SCB by employing a wavelet neural network?WNN?model based on the data characteristics of SCB.To be specific,two SCB values of adjacent epoch subtract each other to get the corresponding single difference sequence of SCB,and then,the sequence is preprocessed through using the preprocessing method designed for the single difference sequence.The subsequent step is to model the WNN based on the preprocessed sequence.At the same time,a genetic algorithm is used to optimize the initial network parameters of WNN in the process of modeling.After the WNN model is determined,the single difference values are predicted based on time series.Lastly the predicted single difference values are restored to the corresponding predicted SCB values.The simulation results have shown that the proposed prediction principle based on the single difference sequence of SCB can make the WNN model architecture simple and the predicting precision higher than that of the general SCB prediction modeling.The designed preprocessing method specific to the single difference of SCB is able to further improve the prediction performance of the WNN model by reducing the effect from outliers.The proposed SCB prediction model outperforms the IGU-P solutions at least on a daily basis,and its medium-and long-term prediction performance is also better than that of several frequently-used SCB models. |
PDF Full Text Request