Optimizing network RTK positioning accuracy with centralized Kalman filtering

Abstract

Real-time kinematic (RTK) and network RTK techniques have been widely used to achieve centimeter-level high-precision positioning. The virtual reference station concept is a practically efficient approach to obtain the benefits of using multiple reference stations. However, under this concept, the full covariance information of the virtual measurement errors is not provided to users, leading to suboptimal positioning performance. Toward RTK applications where communicational and computational resources are rich and accuracy performance is sensitive, this paper proposes an undifferenced, uncombined centralized RTK positioning framework that maximizes the benefits from using multiple reference stations. This approach produces an optimal positioning solution under Gaussian noise with correctly-specified error covariances. Monte Carlo simulations show that the performance gain of the proposed method increases with network scale. At a 50 km baseline, it reduces the vertical RMS error by 48% relative to the conventional approach. Additional simulations demonstrate stable performance under ionospheric covariance-model mismatch and irregular reference-station geometry. Real-data experiments covering compact networks, dynamic operation, multiple users, and longer baselines consistently confirm improved positioning accuracy, with vertical RMS reductions ranging from 23 to 90%. The experimental results also suggest that the proposed approach enables a faster ambiguity fix than conventional approaches. These results demonstrate the accuracy benefits of centralized multi-reference-station RTK processing.