Estimation of GPS L2 signal observables using Multilayer Perceptron Artificial Neural Network for positional accuracy improvement

In recent decades, due to the increasing mobility of people and goods, the rapid growth of users of mobile devices with location-based services has increased the need for geospatial information. In this context, positioning using data collected by the Global Navigation Satellite Systems (multi-GNSS) has gained more importance in the field of geomatics. The quality of the solutions is related, among other factors, to the receiver's type used in the work. To improve the positioning with low-cost devices and to avoid additional user expenses, this work aims to propose the implementation of an Artificial Neural Network (ANN) to estimate the GPS L2 carrier observables. For this, a network model was selected through the cross-validation (CV) technique, the observations were estimated, and the accuracy of the solutions was analyzed. The CV technique demonstrated that a Multilayer Perceptron with four intermediate layers and one with one intermediate layer are the most appropriate configurations for this problem. The dual-frequency RINEX processing (with artificial data) revealed significant improvements. For some tests, it was possible to comply with the rural property georeferencing regulations of the Brazilian National Institute of Colonization and Agrarian Reform (INCRA). The results indicate, therefore, that the methodological proposal of the present investigation is very promising for approximating the quality of positioning reachable using a dual-frequency receiver.