GPS processing, like every processing method for geodetic applications, relies upon least-squares estimation. Quality measures must be defined to assure that the estimates are close to reality. These quality measures are reliable provided that, first, the covariance matrix of the observations (the stochastic model) is well defined and second, the systematic effects are completely removed (i.e., the functional model is good).

In the GPS precise point positioning (PPP) the stochastic and functional models are not as complicated as in the differential GPS processing. We will assess the quality of the GPS Precise Point Positioning in this thesis.

To refine the functional model from systematic errors, we have 1) used the phase observations to prevent introducing any hardware bias to the observation equations, 2) corrected observations for all systematic effects with amplitudes of more than 1cm, 3) used undifferenced observations to prevent having complications (e.g. linearly related parameters) in the system of observation equations.

To have a realistic covariance matrix for the observations we have incorporated the ephemeris’ uncertainties into the system of observation equations.

The above-mentioned technique is numerically tested on the real data of some of the International GNSS Service stations. The results confirm that undifferenced stochastic-related properties (e.g. degrees of freedom) can be reliable means to recognize the parameterization problem in differenced observation equations. These results also imply that incorporation of the satellite ephemeris uncertainties might improve the estimates of the station positions.