Time Series Analysis of GPS Observables

Today, GPS receivers for positioning and timing are capable to output range observations at high rates, typically once per second or even higher, for instance at 5 or 10 Hz. One is easily provided with dense and extensive time series of code and phase observations to all GPS satellites in view, on both the L1 and L2 frequency. The observations’ noise is commonly assumed to be white, i.e. that consecutive observations are not correlated. Neglecting however any severe time correlation leads to significant consequences on the results of the data processing, that is, on the quality of the parameters of interest. Therefore in this contribution the time behaviour of the GPS observations’ noise is explored using standard techniques from modern time series analysis. Two different mathematical models for processing are employed and data from zero baseline experiments are used. The analyses show that the noise may not always be independent between consecutive observables. In particular at higher sampling rates, time correlation does enter the scene. Correlation between neighbouring observations turns out to be significant for 5 Hz data; the correlation coefficient may reach a value of 0.8 to 0.9. An attempt has been made to describe the time correlation by a first order auto-regressive noise process.