A Random Forest-Based Method for Differential Correction Compensation in Cooperative Positioning

Dynamic base station (DBS) differential Global Navigation Satellite System (DDGNSS) is an extension of the DGNSS. In urban canyons, the use of DBSs holds the potential to enhance the efficiency of positioning systems and save computing and infrastructure costs. However, the positioning uncertainty of DBSs may cause errors in their differential corrections, leading to a degradation in positioning performance. In this article, we propose a random forest (RF)-based compensation method for DBSs. This method utilizes the satellite observation data from DBSs to construct a dataset for model training. The input features include satellite elevation, satellite azimuth, carrier-to-noise ratio, and pseudorange residual sum of squares (PRSSs). The trained model subsequently predicts the error of the differential corrections and compensates for these errors, thus enhancing the accuracy of the DBS-aided differential positioning. A single DBS may encounter issues such as insufficient accuracy of satellite observations and communication interruptions with the mobile station. The reliability of differential correction positioning can be enhanced by jointly utilizing multiple DBSs. A field test conducted in a challenging urban canyon environment demonstrates the efficiency of the proposed method in significantly reducing errors in DBS differential corrections. Besides, cooperative positioning involving DBSs exhibits superior accuracy, when compared to positioning using a single base station.