PRIDE PPP-AR Version 3.2 with AI Integer Ambiguity Validation released

On October 20, 2025, the PRIDE team at the State Key Laboratory of Precision Geodesy, Innovation Academy for Precision Measurement Science and Technology, CAS, released version 3.2 of PRIDE PPP-AR (https://github.com/PrideLab/PRIDE-PPPAR), which supports AI-based validation for undifferenced ambiguity resolution. This version integrates an XGBoost-based machine learning model for ambiguity validation through multi-dimensional indicators, including R-Ratio, Difference Test, Project Test, W-Ratio, ADOP, and the penalty factor λ. Combined with the LAMBDA method, it is designed for processing datasets with durations shorter than six hours and supports both single-system and multi-GNSS static and kinematic positioning. In tests conducted over the four years using stations independent of the training stations, this ML model consistently outperformed the classical Ratio Test, achieving a 5–10% improvement in validation accuracy (as shown below).

Test result

 1. Module Introduction

 Core Upgrade Module: arsig Module: The arsig module is the core upgrade in PRIDE PPP-AR v3.2, responsible for SDBS (single-difference between satellites) and ambiguity resolution, with a newly added AI-based integer ambiguity validation function.

 Basic Functionality: Utilizes SDBS to eliminate receiver-end hardware delays, handles wide-lane and narrow-lane ambiguities separately. Wide-lane ambiguities are uniformly resolved using integer rounding; narrow-lane ambiguities employ different resolution strategies based on data processing duration.

 Duration Adaptation Strategy: For processing duration > 6 hours, integer rounding is applied. For duration ≤ 6 hours, the default approach first reduces correlation using the LAMBDA method, followed by validation of fixed results via the XGBoost model.

 The core data processing flow of PRIDE PPP-AR 3.2 remains consistent with previous versions. the key enhancement is the integration of an AI validation step in the "ambiguity resolution phase" following the LAMBDA algorithm. The detailed workflow is illustrated below:

 Upon entering the ambiguity resolution stage:

 (1) Read base data and configurations: Load "amb_" files, "neq_" files, configuration files, etc., to obtain ambiguity-related settings while retrieving undifferenced ambiguity real-number solutions.

 (2) Wide-lane ambiguity processing: Calculate the SDBS wide-lane ambiguity estimate, then fix the SDBS wide-lane ambiguity using integer rounding.

 (3) Initial narrow-lane ambiguity calculation: Compute the SDBS narrow-lane ambiguity estimate.

 (4) Narrow-lane ambiguity resolution strategy selection:

 Rounding Method: Directly fix the SDBS narrow-lane ambiguity using integer rounding, then count the independent ambiguities that can be fixed for both wide-lane and narrow-lane ambiguities, and output them to the "cst_" file.

 LAMBDA+AI validation Method: Map the undifferenced covariance matrix to the SDBS correlation factor matrix, then use the LAMBDA method for fixing. Calculate parameters such as Ratio and ADOP values and input them into the XGBoost model for validation. If validation fails, reduce the number of candidate ambiguities and repeat the LAMBDA fixing process until validation succeeds. After successful validation, count the independent ambiguities and output them to the "cst_" file. Finally, incorporate strong constraints to output the fixed solution.

2. Key Parameter Configuration

 (1) Ambiguity Resolution Method Parameter (-x <num> or --fix-method <num>)

 This parameter specifies the concrete method for ambiguity resolution, offering three options. By default, it automatically adapts based on processing duration:

 1) Rounding Method: Directly fixes ambiguity by rounding, suitable for scenarios with longer processing times.

 2) LAMBDA+Ratio validation Method: Reduces correlation using the LAMBDA method and validates the fixed result via the Ratio value (default validation threshold is 3).

 3) LAMBDA + AI Validation Method: Reduces correlation using the Lambda method and validates the fixed result through an XGBoost model.

 Default Rules: For processing duration under 6 hours, automatically applies Method 3) Lambda+AI Validation; Ratio Test optional (user-specified). For processing duration ≥ 6 hours, automatically applies Method 1) Rounding Method.

 (2)AI Ambiguity Validation Switch in Configuration File

 This configuration controls whether AI validation is enabled, interacting with the "--fix-method" parameter:

 Value "YES": Enables XGBoost model validation for ambiguity resolution, corresponding to "--fix-method 3".

 Value "NO": Disables AI validation. Uses the Ratio value to determine fixed results, corresponding to "--fix-method 2". The specific Ratio validation threshold is determined by the "Critical search" configuration option.

References

Guo, J., Geng, J. Integer ambiguity validation through machine learning for precise point positioning. Satell Navig 6, 14 (2025). https://doi.org/10.1186/s43020-025-00167-8