Research On The Navigation Satellite Clock Error Prediction Based On The Combination Of Polynomial Model And Functional Network

Satellite clock error is one of the main error sources in precision positioning of the navigationsatellite. Establishing accurate clock error prediction model has considerably practical significance forimproving the accuracy of satellite positioning and enhancing the independent operation capability ofthe navigation constellation. This paper emphasizes on the method which can improve the accuracy ofnavigation satellite clock error prediction. The main works and conclusions are summarized asfollows:1. It is shown that the clock error can be separated by a deterministic part and a random one byanalyzing its data. The former can be quantified by the addition of a constant initial time offset,constant frequency offset and constant frequency drift; the latter is characterized by five independentpower law noises. This paper summarizes the methods of testing and evaluating the frequencycharacteristics of the atomic clock in time domain. At last GPS satellite clocks are taken for example,analyzing their frequency characteristics by using the modified Allan variance.2. Detailed comparision are given of several existing traditional methods of navigation satelliteclock error prediction, including the polynomial model, spectral analysis model, the grey model,ARIMA model and Kalman filtering model. Taking the GPS satellites for example, the simulationresults of each method are shown and the advantages and disadvantages have also been discussed.3. Then this paper introduces the basic theory of functional network, including basic elements,topology and learning process. The Henon map example shows that functional network archetechturesprovide simple and efficient techniques to model nonlinear time series.4. Finally this paper proposes a new satellite clock error prediction method based on thecombination of polynomial model and functional network. According to data characteristics of atomicclock, the method fits the clock error series by polynomial model firstly, and then models thepolynomial fitting residual based on functional network whose structure is defined by waveletde-noising and phase reconstruction. Finally the GPS satellites are taken for example, with both shortand long-term prediction. The simulation results show that the proposed method can fit and predict theclock error series effectively.