Non-ideal Phase Error Compensation Based on Global-Aware 3-D U-Net for GNSS-InSAR

GNSS-InSAR employs ground-based fixed receivers to simultaneously capture scene reflection signals from multiple navigation satellites, enabling three-dimensional deformation monitoring. However, asymmetric multi-static configurations introduce various non-ideal interferometric phase errors, while the system’s relatively SNR and resolution further exacerbate the challenges of error compensation. This paper proposes a GNSS-InSAR non-ideal phase error compensation method based on a globally aware three-dimensional U-Net. To overcome the constraint of scarce training data, a simulated dataset was generated by modelling deformation processes through four temporal patterns and a six-parameter spatial model. Various non-ideal errors were incorporated into the dataset based on the constructed signal model. To achieve spatio-temporal coherence across the entire scene, the network incorporates a global-aware module, employs a 2D-3D hybrid encoding-decoding mechanism, and adopts an alternating training strategy that balances local extremum points with global accuracy. Experimental validation was conducted across natural slopes and highway bridges covered by eight BeiDou IGSO satellites. Measurement accuracy was assessed using differential GNSS positioning data, demonstrating the network’s exceptional performance within GNSS-InSAR systems. This study pioneers the application of a globally aware 3D U-Net for GNSS-InSAR phase error compensation, significantly enhancing three-dimensional deformation monitoring accuracy. It outperforms conventional methods in both landslide and bridge field data, demonstrating strong practical utility and scalability potential.