An Atmospheric Data-Driven Q-Band Satellite Channel Model With Feature Selection

This article proposes a novel atmospheric data-driven Q-band satellite channel model using two artificial neural networks, i.e., multilayer perceptron and long short-term memory (LSTM), to estimate real-time channel attenuation at Q-band via a set of atmospheric parameters. Seven atmospheric parameters for modeling satellite channel attenuation are selected by the least absolute shrinkage and selection operator (LASSO) algorithm from 14 commonly used atmospheric parameters. Simulation results demonstrate that the multilayer perceptron-based atmospheric data-driven Q-band satellite channel model via those seven atmospheric parameters is more accurate and less complex than that via the 14 atmospheric parameters. Meanwhile, the accuracy performance of multilayer perceptron- and LSTM-based atmospheric data-driven Q-band satellite channel models, such as absolute errors and mean-squared errors (MSEs), is discussed and analyzed. The complexity of multilayer perceptron and LSTM in this model, such as training time, loading time, and estimation time, is also investigated. It can be seen that the estimated channel attenuation can well align with the measured channel attenuation.

Automated summary

This article proposes a novel atmospheric data-driven Q-band satellite channel model using two artificial neural networks, multilayer perceptron and long short-term memory (LSTM), to estimate real-time channel attenuation. By selecting seven atmospheric parameters, the model can accurately estimate satellite channel attenuation and is less complex compared to a model using 14 parameters. The accuracy performance and complexity of the multilayer perceptron and LSTM in this model are also analyzed.