Artificial neural network for detecting incorrectly fixed phase ambiguities for L1 mass-market receivers

One of the main challenges in global navigation satellite systems (GNSS) network real-time kinematic positioning is phase ambiguity estimation. We describe methods that predict false fixing (FF) of phase ambiguities in mass-market receivers. In this work, FF is defined to occur when the differences between 3D coordinates estimated in real-time differ by more than 20 cm with respect to the reference coordinates. Phase ambiguity FF events occur for many reasons, such as wrong estimation of phase ambiguities by the network software, noise in the corrections, and the environment of the rover. Moreover, one of the main reasons for phase ambiguity FF is the high level of noise and the low redundancy of observation by receivers that track L1 frequencies only. We develop and analyze a specific tool utilizing an artificial neural network that, when trained, tested, and refined specifically for GNSS mass-market receivers, can predict and detect FF. This tool comprises three inputs for all epochs, the Horizontal Dilution of Precision index, the latency of the differential correction, and the number of satellites with fixed phase ambiguities seen by the rover. It provides as output an index consisting of values 0 or 1, i.e., 0 for no FF and 1 for FF. A description of the training and validating phases is provided. The results of tests show that the algorithm has a 99.7% probability of detecting phase ambiguity FF in these cases.