Ionospheric scintillation simulation based on neural networks.

Thereis a demand for the development of GNSS positioning processing techniques that are more tolerant to the effects of the low latitude ionosphere (in particular, scintillation). The possibility of simulating scintillating channels supports the development of more sophisticated test benches and receivers. This paper proposes a neural network-based simulator of ionospheric amplitude scintillation. This synthetic scintillation simulator uses autoencoders and generative adversarial networks (GANs) to generate time series that follow the statistical characteristics of the $\alpha-\mu$ fading model. A part of the proposed network tries to create a synthetic signal, similar to the field data. The proposed neural network was trained and validated with scintillation data acquired in Sao Jose dos Campos, Brazil, in February 2012 and November 2014. The results of the proposed method show that the simulator yields the correct values of the scintillation index, and the estimated fading coefficients are also close to the specified values. These aspects show that this kind of approach can be promising in the simulation of fading channels. Future improvements of the model are also be discussed.