A CNN-Based Type Recognition Method for Outliers of BDS Satellite Clock Bias

  Objectives  There are many different types of outliers in the satellite clock bias of BeiDou satellite navigation system(BDS), and these outliers directly affect the prediction quality of clock bias and the reliability of performance analysis. The preprocessing of clock bias is the precondition for the prediction and analysis of clock bias, and the recognition of outliers is the key to the preprocessing.

  Methods  This paper proposes a convolutional neural network(CNN)-based recognition method for outliers of BDS satellite clock bias. Firstly, a strategy is proposed to identify outliers by graphical feature differences of clock bias. Based on the principle of image recognition, outliers with different graphic features are classified to make the judgment of outliers more consistent with human cognitive habits. Then, a CNN-based recognition method for BDS clock bias is designed. Finally, by optimizing the model parameters, the accuracy of model recognition is improved continuously.

  Results  By using the BDS satellite clock data publicly provided by the International GNSS Service (IGS) data center, a training set and two test sets containing both phase data and frequency data are made. Under the same experimental conditions, the accuracy of phase model and frequency model in these two test sets are more than 99.6%.

  Conclusions  The proposed method can accurately identify the types of outliers and improve the quality of the pretreatment of clock bias.