Optimizing the Use of GPS Multi-Reference Stations for Kinematic Positioning

Abstract

Differential GPS is able to provide cm-level positioning accuracies, as long as the carrier phase ambiguities are resolved to integer values. Classical methods are based on the use of a single reference station located in the vicinity of the rover. Due to the spatial decorrelation of the errors, the distance between the reference station and the user is generally limited to within 20-30 km or even less, mainly due to the ionosphere. The Multi-Ref method, developed at the University of Calgary, uses a network of reference stations to generate regional code and carrier phase corrections, which can be transmitted to users in order to increase the distance over which integer ambiguity resolution is possible. In the original method, the correlated errors, due to the satellite orbits, troposphere, and ionosphere are modeled together using the L1 and widelane observables. In this paper, extensive efforts were carried out towards optimizing the Multi-Ref method with the objective of maximizing its performance. Data collected in Southeastern Brazil was used in this research. At first, the impact of using covariance functions calculated with different data sets was assessed, showing improvement variations of up to 15% in the observation domain compared to using no network corrections, with the exact improvement depending on the data set used in the computation. A new approach, also using Least-Squares collocation, was then proposed to separately model the correlated errors. An additional effort was carried out in terms of modeling the ionosphere into directional components. Results of the enhanced method showed the same level of improvement as those obtained using the original covariance functions. However, this new approach has advantages with respect to the transmission of the corrections. Finally, an additional step was taken in terms of applying a Kalman filter to the corrections in order to improve their quality. For cases when the corresponding satellite was setting at low elevations, the filter approach improved results up to 44%. A study on the impact of the various covariance functions on the estimated accuracy of the corrections is also included.