Clock Bias Gross Error Detection Method Based on WS-MAD

Abstract

This paper proposes a clock bias gross error detection method, which combines weight- ed support vector data description (SVDD) with modified median absolute deviation (MAD) to address the limitations of the traditional MAD method. The method uses the local reachable density as the weighting factor of SVDD, which fully considers the local characteristics of each data. By constructing the minimum hypersphere in the high-dimen- sional feature space, the one-dimensional data are transformed into the distance from the vector point to the center of the hypersphere in the high-dimensional space. This trans- formation increases the discrepancy between normal data and gross errors. The modified MAD method probes the distance from the current epoch vector point to the center of the hypersphere for gross errors, and thus determines whether the current clock bias is a gross error. The precision clock bias data of BDS-3 with different sampling intervals provided by GFZ were used for simulation experiments. By comparing the results of the MAD method and the WS-MAD method, it is found that the WS-MAD method can detect the small gross errors in the smooth clock bias frequency data and more gross errors in the clock bias data with trend term floating. The fitting and prediction analyses on satellites with different orbits and clock types show that the WS-MAD method improves the fitting and prediction accuracy of MEO and IGSO satellites better than that of GEO satellites. For the same MEO satellites, the enhancement effect of hydrogen clocks is better than that of rubidium clocks.