The parameters retrieved from GPS occultation data can be assimilated into numerical weather prediction model. The accuracy of excess phase is affected by the noise from ionosphere and L2 frequency, especially, when L2 signal is encrypted. In order to improve the accuracy of excess phase and remove the noise by the single Difference (SD) method from reference link into occulted link, we use the updated single difference (USD) method to compute atmospheric excess phase. The combined frequency LC data from the reference link are used and smoothed in an interval time window of 2 s. The antenna phase center correction and relativity correction are estimated. The USD method is validated by utilizing COSMIC data. The case analyses of representative rising and setting occultation events indicate that the excess phases from USD and COSMIC are in good agreement. The excess phase of globally distributed 440 occultation events is also computed. The results show that USD method can obviously improve the distribution of mean bias and standard deviation of bias, which demonstrates the reliability of our USD algorithm. The USD method can be applied to the data processing for FY-3 GNOS (GNss Occultation Sounder) occultation mission which uses semi-codeless technique to track L2 signal.