DeepTCAR: deep learning enhanced three-carrier ambiguity resolution for BeiDou RTK in urban environments

The BeiDou Global Navigation Satellite System (BDS) provides multi-frequency signals that facilitate carrier phase Ambiguity Resolution (AR). The widely known Three-Carrier Ambiguity Resolution (TCAR) method is proven to achieve rapid AR under open-sky conditions. However, in complex urban environments, signal blockages from buildings and multipath effects induced by Non-Line-of-Sight (NLOS) propagation contribute to significant pseudorange errors, which severely degrade TCAR performance and overall positioning accuracy. To address this challenge, an improved Multi-Layer Perceptron (MLP) model is developed to directly predict and correct Double-Differenced (DD) pseudorange errors. The corrected observations are integrated into the TCAR framework to enhance ambiguity resolution performance in challenging urban environments. Results from urban vehicular experiments demonstrate that the proposed method significantly outperforms conventional TCAR, achieving an average improvement of 41.2% in ambiguity fixing success rate across the test scenarios.