In global positioning system (GPS) data processing, incorrect stochast
ic models for double-differenced measurements will result in unreliabl
e statistics for ambiguity search and biased positioning results. In t
he commonly used stochastic model, it is usually assumed that all the
raw GPS measurements are independent and that they have the same varia
nce. In fact,these assumptions are not realistic. Measurements obtaine
d from different satellites cannot have the same accuracy due to varyi
ng noise levels. In this paper, a new method based on modern statistic
al theory is proposed to directly estimate the covariance matrix for d
ouble-differenced GPS measurements. Three different stochastic models
have been tested and analyzed. Test results indicate that by using the
proposed stochastic models the volume of ambiguity search space can b
e reduced and the reliability of the ambiguity resolution is improved.
Also, the statistics of the baseline components estimated with the pr
oposed stochastic models are more efficient.